Related references
Note: Only part of the references are listed.
Article
Materials Science, Multidisciplinary
Enzo Marino et al.
Summary: We extended the EUCLID method for automated material model discovery and identification to linear viscoelasticity by adopting a generalized Maxwell model expressed by a Prony series. The strategy consists of two stages: Stage 1 uses sparse regression to enforce momentum balance and reduce the number of terms in the Prony series, while Stage 2 employs k-means clustering to further reduce the terms by grouping together Maxwell elements with similar relaxation times. This approach was successfully demonstrated on artificial numerical data, accurately identifying a linear viscoelastic model with five relaxation times from a library with hundreds of terms spanning seven orders of magnitude.
MECHANICS OF MATERIALS
(2023)
Article
Mathematics, Applied
Solve Eidnes et al.
Summary: Hybrid machine learning based on Hamiltonian formulations has been successfully demonstrated on simple mechanical systems. A pseudo-Hamiltonian formulation is introduced to generalize the Hamiltonian formulation via the port-Hamiltonian formulation, allowing pseudo-Hamiltonian neural network models to learn external forces acting on a system. Numerical results and an improved training scheme are provided for different systems.
PHYSICA D-NONLINEAR PHENOMENA
(2023)
Article
Engineering, Multidisciplinary
Moritz Flaschel et al.
Summary: We extend our unsupervised automated discovery approach (EUCLID) to unknown constitutive behavior, covering a wide range of important classes. This approach leverages the theory of generalized standard materials to automatically discover the two scalar thermodynamic potentials that define their behavior. The enforced convexity constraint ensures stability and consistency, and sparsity promoting regularization leads to a simple and interpretable model with few internal variables.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Engineering, Multidisciplinary
Beatriz Moya et al.
Summary: Digital twins are digital representations of physical entities that use real-time data to understand the operating conditions. This article presents a unique type of digital twin that combines computer vision, scientific machine learning, and augmented reality. This digital twin can observe and interpret what it sees, make adjustments to its model when necessary, and present the information in augmented reality.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Ravi G. Patel et al.
Summary: Using a least squares space-time control volume scheme can more naturally handle regularity requirements, imposition of boundary conditions, entropy compatibility, and conservation, reducing requisite hyperparameters. This framework introduces inductive biases to discover thermodynamically consistent equations of state guaranteeing hyperbolicity, providing a method for discovering continuum shock models.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Nathaniel Trask et al.
Summary: As traditional machine learning tools are increasingly applied to science and engineering applications, physics-informed methods have emerged as effective tools for endowing inferences with properties essential for physical realizability. The approach strengthens physics rules through the exterior calculus framework and learns metric information from data to provide mathematical guarantees for the machine-learned models of physical rules.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Materials Science, Multidisciplinary
Shenglin Huang et al.
Summary: This paper proposes a thermodynamics-based learning strategy for non-equilibrium evolution equations using Onsager's variational principle. The method learns the free energy and dissipation potential from spatio-temporal measurements, and enforces the satisfaction of the second law of thermodynamics. The approach is demonstrated on three physical processes, showing its robustness and versatility.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2022)
Article
Materials Science, Multidisciplinary
Prakash Thakolkaran et al.
Summary: In this study, a new approach for unsupervised learning of hyperelastic constitutive laws using physics-consistent deep neural networks is proposed. The approach uses only displacement and reaction force data, and leverages a physics-motivated loss function based on the conservation of linear momentum to guide the learning process. By employing a specially designed neural network architecture, multiple physical and thermodynamic constraints are automatically satisfied, and the constitutive laws can be accurately learned.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2022)
Article
Engineering, Multidisciplinary
Filippo Masi et al.
Summary: The mechanical behavior of inelastic materials with microstructure is complex and difficult to predict accurately using traditional methods. This paper proposes a Thermodynamics-based Artificial Neural Networks (TANN) approach for modeling such materials. Several examples demonstrate the high accuracy and physical consistency of TANN in predicting macroscopic and microscopic mechanical behavior.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Mathematics, Applied
Salvatore Cuomo et al.
Summary: Physics-Informed Neural Networks (PINN) are a type of neural network that incorporates model equations, such as partial differential equations, as a component. PINNs have been used to solve various types of equations, including fractional equations and stochastic partial differential equations. Current research focuses on optimizing PINN through different aspects, such as activation functions, gradient optimization techniques, neural network structures, and loss function structures. Despite the demonstrated feasibility of PINN in certain cases compared to traditional numerical techniques, there are still unresolved theoretical issues.
JOURNAL OF SCIENTIFIC COMPUTING
(2022)
Article
Multidisciplinary Sciences
Zhen Zhang et al.
Summary: This paper proposes the GENERIC formalism informed neural networks (GFINNs) that obey the symmetric degeneracy conditions of the GENERIC formalism. By leveraging the component-wise architecture design, GFINNs are able to learn the underlying equations and outperform existing methods in simulation problems.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2022)
Article
Physics, Fluids & Plasmas
Marios Mattheakis et al.
Summary: The research introduces a Hamiltonian neural network to solve differential equations governing dynamical systems without using ground truth data, optimizing solely based on predicted functions to improve accuracy. Error analysis indicates numerical errors depend on network performance.
Review
Computer Science, Interdisciplinary Applications
Abel Sancarlos et al.
Summary: This work focuses on the development of a Hybrid Twin model for the new generation of BMS, applying three reduced order model techniques for different ranges of application. Simulation of the electric vehicle system was conducted with real-time performance demonstrated, while a Digital Twin was created for real-time simulations and adaptive predictions. Additionally, a data-driven model based on Dynamic Mode Decomposition techniques was developed to correct the gap between prediction and measurement, creating the first hybrid twin of a Li-ion battery capable of self-correction.
ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING
(2021)
Review
Neurosciences
Fabian Eitel et al.
Summary: This article introduces the application of deep neural networks in medical imaging for more accurate diagnostics and treatment recommendations, and discusses current challenges such as training models on small, biased data sets and algorithmic bias.
EXPERIMENTAL NEUROLOGY
(2021)
Article
Computer Science, Interdisciplinary Applications
David Gonzalez et al.
Summary: A method for data-driven learning of physical phenomena with internal variables and inherent noise is introduced, achieving a description that complies with basic thermodynamic principles.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Quercus Hernandez et al.
Summary: The method developed uses feedforward neural networks to learn physical systems from data while ensuring compliance with the first and second principles of thermodynamics. By enforcing the metriplectic structure of dissipative Hamiltonian systems, it minimizes the amount of data required and naturally achieves conservation of energy and dissipation of entropy in its predictions. No prior knowledge of the system is necessary, and the method can handle both conservative and dissipative, discrete and continuous systems.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Materials Science, Multidisciplinary
Filippo Masi et al.
Summary: TANNs is a physics-based neural network method for constitutive modeling, where thermodynamic principles are encoded into the network architecture using automatic differentiation. Its advantages include efficiency, robustness, accuracy in predictions, and maintaining thermodynamic consistency.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2021)
Article
Thermodynamics
Shengze Cai et al.
Summary: Physics-informed neural networks (PINNs) have gained popularity in engineering fields for their effectiveness in solving realistic problems with noisy data and partially missing physics. Through automatic differentiation to evaluate differential operators and defining a multitask learning problem, PINNs have been applied to various heat transfer problems, bridging the gap between computational and experimental heat transfer.
JOURNAL OF HEAT TRANSFER-TRANSACTIONS OF THE ASME
(2021)
Article
Engineering, Multidisciplinary
Quercus Hernandez et al.
Summary: The algorithm presented in this study uses sparse autoencoders to reduce the dimensionality of a physical system and predict its time evolution using deep neural networks. It ensures the conservation of total energy and entropy inequality, making it suitable for both conservative and dissipative systems. The method has been tested in examples from fluid and solid mechanics.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Yunjin Tong et al.
Summary: The researchers proposed a novel neural network architecture Taylor-nets for long-term prediction of Hamiltonian dynamic systems while preserving their symplectic structures. The algorithm performs well in terms of prediction accuracy, convergence rate, and robustness, even with extremely small training data and a short training period.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Multidisciplinary Sciences
John Jumper et al.
Summary: Proteins are essential for life, and accurate prediction of their structures is a crucial research problem. Current experimental methods are time-consuming, highlighting the need for accurate computational approaches to address the gap in structural coverage. Despite recent progress, existing methods fall short of atomic accuracy in protein structure prediction.
Proceedings Paper
Automation & Control Systems
Eric Heiden et al.
Summary: Differentiable simulators offer a way to close the gap between simulation and reality by using efficient, gradient-based optimization algorithms. The augmentation of the novel differentiable physics engine with neural networks allows for learning nonlinear relationships and modeling effects not considered by traditional simulators. This hybrid simulation approach requires less data for training and shows improved generalization compared to entirely data-driven models.
2021 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2021)
(2021)
Article
Physics, Fluids & Plasmas
Shaan A. Desai et al.
Summary: Accurately learning the temporal behavior of dynamical systems requires models with well-chosen learning biases. Recent innovations embed the Hamiltonian and Lagrangian formalisms into neural networks, demonstrating significant improvement in predicting trajectories of physical systems. Despite the success of this approach, real world dynamical systems often experience energy dissipation and are driven by time-dependent forces. This study addresses the challenge of learning from such nonautonomous systems by embedding the port-Hamiltonian formalism into neural networks, showing efficient learning and prediction capabilities for chaotic systems.
Article
Physics, Multidisciplinary
Chen-Di Han et al.
Summary: The increasing interest in using Hamiltonian neural networks to predict chaotic systems has led to the development of a new class of HNN capable of adaptively predicting nonlinear physical systems. By training the network on time series from a small number of parameter values, the HNN can accurately forecast the system's state across a wide parameter range.
PHYSICAL REVIEW RESEARCH
(2021)
Article
Computer Science, Interdisciplinary Applications
Ameya D. Jagtap et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2020)
Review
Computer Science, Interdisciplinary Applications
Francisco Chinesta et al.
ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING
(2020)
Article
Engineering, Multidisciplinary
Zhiping Mao et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Mathematics, Applied
S. Conti et al.
ARCHIVE FOR RATIONAL MECHANICS AND ANALYSIS
(2020)
Article
Chemistry, Physical
David Gonzalez et al.
Article
Multidisciplinary Sciences
M. Grmela et al.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2020)
Review
Automation & Control Systems
Ramy Rashad et al.
IMA JOURNAL OF MATHEMATICAL CONTROL AND INFORMATION
(2020)
Article
Multidisciplinary Sciences
Beatriz Moya et al.
Review
Mathematics, Applied
Karim Cherifi
PHYSICA D-NONLINEAR PHENOMENA
(2020)
Article
Computer Science, Artificial Intelligence
Pengzhan Jin et al.
Article
Physics, Mathematical
E. Weinan
COMMUNICATIONS IN COMPUTATIONAL PHYSICS
(2020)
Article
Physics, Fluids & Plasmas
Anshul Choudhary et al.
Article
Thermodynamics
David Gonzalez et al.
CONTINUUM MECHANICS AND THERMODYNAMICS
(2019)
Article
Automation & Control Systems
Christopher A. Beattie et al.
Article
Computer Science, Interdisciplinary Applications
M. Raissi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Article
Materials Science, Multidisciplinary
David Gonzalez et al.
FRONTIERS IN MATERIALS
(2019)
Article
Mathematics, Interdisciplinary Applications
B. Moya et al.
COMPUTATIONAL MECHANICS
(2019)
Article
Mathematics, Applied
Tom Bertalan et al.
Article
Physics, Mathematical
Chao Ma et al.
COMMUNICATIONS IN COMPUTATIONAL PHYSICS
(2019)
Article
Mathematics, Applied
Guofei Pang et al.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2019)
Article
Astronomy & Astrophysics
Takahiro Nishimichi et al.
ASTROPHYSICAL JOURNAL
(2019)
Review
Computer Science, Interdisciplinary Applications
Ruben Ibanez et al.
ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING
(2018)
Article
Mathematics, Applied
S. Conti et al.
ARCHIVE FOR RATIONAL MECHANICS AND ANALYSIS
(2018)
Review
Physics, Multidisciplinary
Miroslav Grmela
JOURNAL OF PHYSICS COMMUNICATIONS
(2018)
Proceedings Paper
Biochemistry & Molecular Biology
Marino Arroyo et al.
ROLE OF MECHANICS IN THE STUDY OF LIPID BILAYERS
(2018)
Article
Chemistry, Physical
Pep Espanol et al.
JOURNAL OF CHEMICAL PHYSICS
(2017)
Article
Mathematics, Interdisciplinary Applications
Ruben Ibanez et al.
COMPUTATIONAL MECHANICS
(2017)
Article
Mathematics
E. Weinan
COMMUNICATIONS IN MATHEMATICS AND STATISTICS
(2017)
Article
Engineering, Multidisciplinary
T. Kirchdoerfer et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2016)
Article
Mathematics, Applied
Bruce Turkington et al.
Review
Computer Science, Artificial Intelligence
G. Peter Zhang
IEEE TRANSACTIONS ON SYSTEMS MAN AND CYBERNETICS PART C-APPLICATIONS AND REVIEWS
(2007)
Article
Physics, Multidisciplinary
M Grmela
Article
Engineering, Electrical & Electronic
SC Lee et al.
ELECTRIC POWER SYSTEMS RESEARCH
(2002)