Physics-informed Neural Networks (PINN) for computational solid mechanics: Numerical frameworks and applications

Thin-Walled Structures, Год журнала: 2024, Номер 205, С. 112495 - 112495

Опубликована: Сен. 24, 2024

Язык: Английский

---

From PINNs to PIKANs: recent advances in physics-informed machine learning

Machine learning for computational science and engineering, Год журнала: 2025, Номер 1(1)

Опубликована: Март 11, 2025

Язык: Английский

---

An enhanced hybrid adaptive physics-informed neural network for forward and inverse PDE problems

Applied Intelligence, Год журнала: 2025, Номер 55(4)

Опубликована: Янв. 2, 2025

Язык: Английский

---

f-PICNN: A physics-informed convolutional neural network for partial differential equations with space-time domain

Journal of Computational Physics, Год журнала: 2024, Номер 515, С. 113284 - 113284

Опубликована: Июль 15, 2024

The physics and interdisciplinary problems in science engineering are mainly described as partial differential equations (PDEs). Recently, a novel method using physics-informed neural networks (PINNs) to solve PDEs by employing deep with physical constraints data-driven models has been pioneered for surrogate modelling inverse problems. However, the original PINNs based on fully connected pose intrinsic limitations poor performance nonlinearity, drastic gradients, multiscale characteristics or high dimensionality which complex features hard capture. This leads difficulties convergence correct solutions computational costs. To address above problems, this paper, convolutional network framework finite discretization schemes stack of series nonlinear units (NCUs) solving space-time domain without any labelled data (f-PICNN) is proposed, memory mechanism can considerably speed up convergence. Specifically, initial conditions (ICs) hard-encoded into first time-step solution used extrapolate next solution. Dirichlet boundary (BCs) constrained soft BC enforcement while Neumann BCs enforced. Furthermore, loss function designed set discretized PDE residuals optimized conform laws. Finally, proposed auto-regressive model proven be effective wide range 1D 2D both space time under different (e.g., Euler, Crank Nicolson fourth-order Runge-Kutta). numerical results demonstrate that not only shows ability learn efficiently but also provides an opportunity greater conceptual simplicity, potential extrapolation from learning limited dataset.

Язык: Английский

---

Predicting and analyzing the three-dimensional spatiotemporal evolution process of tidal currents by using a brand new machine learning algorithm

Physics of Fluids, Год журнала: 2025, Номер 37(1)

Опубликована: Янв. 1, 2025

A machine learning algorithm was developed for efficiently predicting the 3D (three-dimensional) spatiotemporal evolution process of tidal currents and analyzing their spatial distribution characteristics. In algorithm, an extremely simplified multi-layer perceptron architecture, embedded information method, a splicing-sharing method at different water depths were used to achieve high-coverage, comprehensive, systematic current prediction study area. The can predict future time series three-dimensional movement solves problem that existing algorithms are unable analyze similarity over many years. this study, evolutions in southern waters Liaoning Province, China, analyzed. Finite-Volume Coastal Ocean Model ocean model simulate zone, generating dataset train model. trained then currents. results show has high accuracy period 12 h, with R2 (R-Square) 0.871, mean absolute error 0.047 m/s root square 0.152 m/s. Additionally, could effectively correlation characteristics depths, similar processes zones also be classified.

Язык: Английский

---

Inversion of Multiple Reservoir Parameters Based on Deep Neural Networks Guided by Lagrange Multipliers

SPE Journal, Год журнала: 2025, Номер unknown, С. 1 - 21

Опубликована: Март 1, 2025

Summary The inversion of reservoir parameters is critically important during oilfield exploration and development, as it provides precise comprehensive information that helps reduce risks by mitigating uncertainties. In this paper, we propose an intelligent method using a Lagrange multipliers-guided physical residual neural network (Lg-PRNN), incorporating nonlinear variations, adaptive parameters, multipliers. use multipliers eliminates the need to manually adjust weights in loss function, significantly improving efficiency. By introducing variations for time space coordinates input, Lg-PRNN can capture changes fluid flow rates, thereby enhancing its ability solve seepage equations under varying conditions. introduced inputs increase flexibility enhance adaptability generalization capabilities. two synthetic experiments one field experiment, permeability, skin factor, wellbore storage coefficient were accurately inverted fitting bottomhole pressure (BHP), demonstrating effectiveness model. Compared with latest methods utilizing networks, not only improves efficiency but also enhances accuracy approximately 72%. Keywords Physical Residual Neural Network, multiplier method, Inversion Deep learning, Numerical well test

Язык: Английский

---

An adaptive multi-scale spatial-temporal graph attention ensemble network with physical guidance for remaining useful life prediction of multi-sensor equipment

Reliability Engineering & System Safety, Год журнала: 2025, Номер unknown, С. 111152 - 111152

Опубликована: Апрель 1, 2025

Язык: Английский

---

Physics-informed neural networks with trainable sinusoidal activation functions for approximating the solutions of the Navier-Stokes equations

Computer Physics Communications, Год журнала: 2025, Номер unknown, С. 109672 - 109672

Опубликована: Май 1, 2025

Язык: Английский

---

An adaptive optimal selection approach of the Mixture-of-Experts model embedded with PINNs for one-dimensional hyperbolic conservation laws

Communications in Nonlinear Science and Numerical Simulation, Год журнала: 2025, Номер unknown, С. 108936 - 108936

Опубликована: Май 1, 2025

Язык: Английский

---

Physics-informed machine learning in geotechnical engineering: a direction paper

Geomechanics and Geoengineering, Год журнала: 2025, Номер unknown, С. 1 - 32

Опубликована: Май 22, 2025

Язык: Английский

---