Differentiable finite element method with Galerkin discretization for fast and accurate inverse analysis of multidimensional heterogeneous engineering structures

Computer Methods in Applied Mechanics and Engineering, Journal Year: 2025, Volume and Issue: 437, P. 117755 - 117755

Published: Jan. 22, 2025

Language: Английский

---

Physics and data hybrid-driven interpretable deep learning for moving force identification

Engineering Structures, Journal Year: 2025, Volume and Issue: 329, P. 119801 - 119801

Published: Feb. 3, 2025

Language: Английский

---

Structural nonlinear boundary condition identification using a hybrid physics data-driven approach

Nonlinear Dynamics, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 23, 2024

Abstract As civil infrastructures often exhibit nonlinearities, the identification of nonlinear behaviors is crucial to assess structural safety state. However, existing physics-driven methods can only estimate parameters given a known behavior pattern. By contrast, data-driven merely map load-response relationship at level, rather than identify an accurate mapping component level. To address these issues, hybrid physics-data-driven strategy developed in this study blind nonlinearity. The components are surrogated by multilayer perceptron, and linear ones simulated using finite element method. Subsequently, global stiffness matrix restoring force vector assembled according elemental topology obtain model. discrepancy between measured model-predicted responses formulated as loss function, minimizing which MLPs indirectly trained nonlinearities be identified without knowing nonlinearity type. Three numerical cases used verify proposed method identifying elastic, hysteretic, multiple boundary conditions. Results show that robust different noise levels, sensor placements, types. Moreover, model possesses strong generalization ability accurately predict full-field responses.

Language: Английский

---

Solving fluid flow in discontinuous heterogeneous porous media and multi-layer strata with interpretable physics-encoded finite element network

Journal of Rock Mechanics and Geotechnical Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 1, 2024

Language: Английский

---

A Comprehensive Investigation of Physics-Informed Learning in Forward and Inverse Analysis of Elastic and Elastoplastic Footing

Computers and Geotechnics, Journal Year: 2025, Volume and Issue: 181, P. 107110 - 107110

Published: Feb. 5, 2025

Language: Английский

---

Explainable artificial intelligence model for the prediction of undrained shear strength

Theoretical and Applied Mechanics Letters, Journal Year: 2025, Volume and Issue: unknown, P. 100578 - 100578

Published: Feb. 1, 2025

Language: Английский

---

State-of-art review on load identification and response reconstruction to realize digital twin of infrastructures

Advances in Structural Engineering, Journal Year: 2025, Volume and Issue: unknown

Published: March 20, 2025

The digital twin (DT) technique for infrastructures has been developed and attracted a significant amount of attention since 2020. Nonetheless, the key technologies used DT, including load identification (LID), response reconstruction (RRE), damage detection, have much longer history than DT itself. By employing these methods, cyber models are established updated to represent operational state real structure, meanwhile, monitored data at discrete locations can be expanded full-field structure realize DT. In this work, LID RRE methods civil under quasi-static dynamic loading actions comprehensively reviewed. LID, four types formulations derived, solutions summarized address inherent ill-posed problems. RRE, model- data-driven reviewed with five levels performance. Subsequently, several sensing techniques introduced. pros, cons, features highlighted. challenges prospects in outline future trend in-service infrastructure digitalization.

Language: Английский

---

Physics-informed neural networks for solving steady-state temperature field in artificial ground freezing

Canadian Geotechnical Journal, Journal Year: 2025, Volume and Issue: 62, P. 1 - 17

Published: Jan. 1, 2025

Artificial ground freezing (AGF) is a widely used technique for soil stabilization and waterproofing. Numerous studies have been devoted to solving the heat transfer problems in AGF while encountering limitations handling complex geometries boundary conditions being computationally intensive. Recently, using machine learning methods predict temperature fields has gained attention, demonstrating potential achieve higher accuracy than conventional models. However, these are typically limited by need large, labeled datasets, which time-consuming difficult obtain. In this study, we address challenges applying physics-informed neural networks (PINNs) solve steady-state problem AGF, focusing on distribution around single pipe. By embedding conduction equation into loss function, PINNs reduce extensive data. To enhance efficiency, employed, results compared against finite element method. Results show that high accuracy, particularly larger domains with moderate gradients, providing competitive performance more configurations involving steeper gradients. This approach offers promising alternative modeling geotechnical applications, implications reducing computational costs design.

Language: Английский

---

The novel graph transformer-based surrogate model for learning physical systems

Computer Methods in Applied Mechanics and Engineering, Journal Year: 2024, Volume and Issue: 432, P. 117410 - 117410

Published: Oct. 2, 2024

Language: Английский

---

Application of a thermo-mechanical-damage coupling model based on the TLF-SPH in the thermal cracking simulation of brittle solids

Computers and Geotechnics, Journal Year: 2024, Volume and Issue: 179, P. 107017 - 107017

Published: Dec. 26, 2024

Language: Английский

---