{"title":"Robust and Automated Tetrahedral Mesh Generation From Imperfect Computer-Aided Design Models for Computational Electromagnetics","authors":"Yifu Wang, Xuejuan Chen, Taoran Liu, Jianjun Chen","doi":"10.1002/nme.70464","DOIUrl":"https://doi.org/10.1002/nme.70464","url":null,"abstract":"<div>\u0000 \u0000 <p>Finite element analysis (FEA) is a fundamental technique in computational electromagnetics (CEM), and its accuracy and stability largely depend on the quality and geometric consistency of the mesh. As modern electronic systems increasingly involve complex computer-aided design (CAD) geometries, conventional meshing methods may fail when the input models contain defects such as self-intersections or overlaps. This paper presents an automated and numerically stable tetrahedral mesh generation framework that can produce solver-ready meshes directly from imperfect CAD models. The framework combines Wild Tetrahedralization for fault-tolerant boundary recovery under floating-point arithmetic, a background-tree-based geometry–mesh mapping for consistent association between mesh elements and CAD entities, and a physics-aware refinement process that couples wavelength and port constraints with adaptive optimization. Experiments on a range of representative electronic models show that the proposed approach reliably generates high-quality, non-degenerate volumetric meshes and maintains stability even for geometries with considerable imperfections.</p>\u0000 </div>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 22","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149391987","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Subwavelength Topology Optimization With Feature-Mapping","authors":"Felix Huber, Fabian Wein, Florian Toth","doi":"10.1002/nme.70458","DOIUrl":"https://doi.org/10.1002/nme.70458","url":null,"abstract":"<p>Acoustic topology optimization has been extensively studied since the earliest developments of structural optimization, and substantial progress has been made in the design of acoustic devices. More recently, wave-focusing applications have attracted growing interest. In particular, sonic black hole (SBH) structures have gained traction for their ability to focus waves with wavelengths exceeding the structure's characteristic size. A key challenge in this context is the need to clearly distinguish between material/void phases. The solid isotropic material with penalization (SIMP) method introduces continuous pseudo-densities, which can lead to intermediate material states that are beneficial for acoustic wave manipulation functions but unphysical. While this issue is often easily handleable in classical applications such as stiffness maximization, it becomes critical for subwavelength acoustic design. Our analysis reveals a strong dependence of transmission characteristics on both frequency and feature size, suggesting that filtering and projlts. In this work, we examine different filtering strategies by optimizing planar SBH-type structures and develop a detailed understanding of acoustic behavior in the regime where the wavelength is significantly greater than the geometric scale. We employ the SIMP method, along with two feature-mapping schemes, to obtain designs that achieve globally optimal performance for both single-frequency and broadband focusing objectives. Finally, we analyze computational efficiency and provide practical recommendations for solving similar classes of acoustic optimization problems.</p>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/nme.70458","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149356700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Modal-Based Framework for Topology Optimization of Compliant Mechanisms With Optimal Actuator Sizing","authors":"Alexander Nowak, Alexander Hasse","doi":"10.1002/nme.70452","DOIUrl":"https://doi.org/10.1002/nme.70452","url":null,"abstract":"<p>The design of compliant mechanisms with integrated actuation requires consistent treatment of structural topology, actuator dimensions, and actuator placement. This work presents a coupled optimization framework based on modal formulations that combines the optimization of structural topology and linear actuators into a unified design approach. The method minimizes actuator volume while ensuring prescribed deformations by explicitly capturing the interaction between modal stiffness and actuator-induced work. A scalable actuator model reflects the proportional dependence of actuator properties on geometric dimensions as exhibited by piezoelectric stack actuators, enabling flexible sizing and placement. The coupling is formulated through a modal stiffness to coordinate the adjustment of structural compliance and actuator effort. Analytical investigations reveal a characteristic dependence of actuator volume on this stiffness, including the existence of a unique volume minimum. Motivated by these findings, the coupled framework combines a modal-synthesis approach for topology design with an analytical procedure for volume-optimal actuator sizing, governed by an outer golden-section search that varies the modal stiffness to minimize the resulting actuator volume. Demonstrated across several case studies – including classical single-input-single-output mechanisms, problems with mechanical amplification requirements, robustness against external loads, and shape-adaptive structures – the proposed framework systematically reduces actuator volume while maintaining the prescribed deformation behavior.</p>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/nme.70452","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149356046","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"From Hyperelastic Physics-Augmented Neural Networks to Fortran User Materials: Standalone Implementation and Computational Performance in Explicit Dynamics","authors":"L. Maurer, S. Eisenträger, D. Juhre, M. Bulla","doi":"10.1002/nme.70439","DOIUrl":"https://doi.org/10.1002/nme.70439","url":null,"abstract":"<p>Data-driven material modeling techniques have gained significant attention due to their ability to capture the complex constitutive behavior beyond the limitations of classical material models. Physics-augmented neural networks (PANNs), which incorporate physical restrictions into their architecture, combine the flexibility of machine learning with properties required for reliable engineering simulations. As a result, they have become increasingly relevant in computational material modeling. This work presents an approach to integrate pretrained isotropic, nearly incompressible, invariant-based hyperelastic PANNs into the explicit finite element solvers <span>Simcenter Radioss</span> and <span>OpenRadioss</span> (Siemens). To this end, a Python framework transfers the trained parameters and the required constitutive expressions of the supported feed-forward architecture to a standalone Fortran user material routine. The analytically expressed first derivatives of the strain-energy density are included in the generated code. The finite element simulation therefore does not require an external machine-learning library or a specialized solver. In addition to the integration strategy, particular emphasis is placed on the computational efficiency of the resulting user material subroutine. The influence of the network architecture on the overall simulation performance is investigated, and approaches for reducing evaluation costs while preserving the quality of the fitted material response are discussed. In this context, replacing the commonly used SoftPlus activation function by the SQuarePlus function reduces the additional constitutive evaluation costs for the investigated networks. To facilitate practical use, a publicly available GitHub repository (\u0000https://github.com/OVGU-CoMe/PANN_Radioss) provides the code generator and the implemented material routines. Within the supported model class, the generator accounts for different numbers and widths of fully connected hidden layers. Additional invariant inputs can be incorporated through localized changes to the input and derivative definitions, whereas different constitutive formulations or solver interfaces require more comprehensive adaptations. As a complex example, an impact simulation demonstrates that the generated PANN user materials can be applied to explicit analyses involving large strains and (self-)contact. The results illustrate the feasibility and computational performance of the investigated hyperelastic PANNs in <span>Simcenter Radioss</span> and <span>OpenRadioss</span>. They also show how a commercial and an open-source finite element solver can be used to study this specified class of machine-learning-based constitutive models.</p>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/nme.70439","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149356291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bin Zhou, Jialong Ding, Shugang Li, Jiang Xu, Haifei Lin, Haiqing Shuang, Shoujian Peng, Liang Cheng, Chongzhi Wang, Mingwei Cui
{"title":"Mechanisms of Geo-Stress and Gas Pressure in Coal and Gas Outburst: Insights From Particle-Fluid Coupled Numerical Simulation","authors":"Bin Zhou, Jialong Ding, Shugang Li, Jiang Xu, Haifei Lin, Haiqing Shuang, Shoujian Peng, Liang Cheng, Chongzhi Wang, Mingwei Cui","doi":"10.1002/nme.70463","DOIUrl":"https://doi.org/10.1002/nme.70463","url":null,"abstract":"<div>\u0000 \u0000 <p>With increasing mining depths, elevated geo-stress and gas pressure substantially raise the risk of coal and gas outbursts (CGO), posing serious threats to mining safety. Conventional approaches including triaxial mechanical experiments, large-scale physical simulations, and existing numerical models have been limited in delineating the distinct roles of geo-stress and gas pressure during the CGO process. In this study, we developed a novel method coupling particle flow with computational fluid dynamics to unravel the underlying CGO mechanisms. Our simulations reveal that geo-stress predominately accelerates damage via tensile and shear cracking, whereas gas pressure exerts a comparatively modest effect on crack propagation. Under coupled geo-stress and gas pressure, the CGO failure zone evolves from a spindle-like to a quarter-ellipsoidal morphology. The internal stress of coal body change appears in four forms: continuous decrease, increase-then-decrease, continuous increase, and periodic decrease–increase. Furthermore, the increase in gas pressure and geo-stress not only leads to higher velocity and greater quantity of outburst coal particles but also accelerates the formation of cracks, promotes earlier coal damage, and premature stress release. These findings provide mechanistic insights into the initiation and evolution of CGO, offering a quantitative basis for the design of targeted prevention strategies in deep mining.</p>\u0000 </div>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149355759","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Aspect-Ratio Preserving Target-Matrix Optimization With Linesweep Smoothing for Arbitrary Lagrangian–Eulerian Meshes With Adaptive Refinement","authors":"A. Colaïtis, S. Guisset, J. Breil","doi":"10.1002/nme.70462","DOIUrl":"https://doi.org/10.1002/nme.70462","url":null,"abstract":"<div>\u0000 \u0000 <p>We present a rezoning method for three-dimensional arbitrary Lagrangian–Eulerian (ALE) and ALE with adaptive mesh refinement (ALE-AMR) simulations. The method combines linesweep smoothing with an aspect-ratio-preserving target-matrix optimization paradigm (TMOP) formulation. It is designed to preserve anisotropic mesh structures produced by the Lagrangian phase while maintaining mesh validity and reducing shear, sliver formation, and cell bunching in block-structured, optionally nonconformal meshes. The approach combines a patch-local TMOP objective and a convexity-based disentanglement metric within an existing linesweep framework. The TMOP target matrix is built from local edge geometry and includes a prescribed aspect-ratio cap, which provides an explicit control parameter for mesh relaxation. The metric is evaluated on an extended nodal patch, so that the objective couples the motion of a node to the quality of neighboring elements affected by that motion. Combined with sequential updates, this ensures that accepted steps are tested against the updated quality of the local coupled patch, rather than only against the elements directly incident on the node. When invalid configurations are detected, the TMOP objective is replaced by a convexity objective that drives signed volumes back to positive values. Parallelism is recovered through a distance-1 coloring of the nodal graph, allowing independent node updates to be performed concurrently. Mesh-only and hydrodynamic tests assess shear reduction, aspect-ratio control, and mesh validity. Among the configurations examined, only the TMOP combined method consistently satisfies these requirements together while retaining anisotropic mesh structures, with no observed invalid post-rezoning states. The prescribed target cap provides explicit control over excessive elongation without imposing an isotropic target. Strong and weak scaling measurements quantify the cost of the parallel optimization.</p>\u0000 </div>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149310883","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Stabilized Quadratic Virtual Element Formulation With Reduced Degrees of Freedom for Two-Dimensional Elasticity","authors":"Peng-Zhao Mo, Ming-Jun Zhao, Song Cen, Yan Shang","doi":"10.1002/nme.70450","DOIUrl":"https://doi.org/10.1002/nme.70450","url":null,"abstract":"<div>\u0000 \u0000 <p>In this work, a stabilized second-order virtual element method is proposed for two-dimensional elasticity problems. Relying solely on low-order degrees of freedom, a Euclidean orthogonal projection in the degree of freedom space is constructed using a least-squares approach. Leveraging this projection, a streamlined serendipity virtual element scheme is established that eliminates redundant moment degrees of freedom within the standard virtual element, achieving a significant reduction in implementation complexity while preserving high accuracy. To validate the performance of the new element, classical numerical examples are presented, with a special emphasis on bending problems of various beam types solved on an extremely coarse mesh—a demanding capability that is scarcely reported in the existing literature on the virtual element method. The proposed element demonstrates superior accuracy in displacement and stress fields, achieving optimal convergence rates in both the <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <msup>\u0000 <mi>L</mi>\u0000 <mn>2</mn>\u0000 </msup>\u0000 </mrow>\u0000 <annotation>$$ {L}^2 $$</annotation>\u0000 </semantics></math> norm and energy semi-norm. Numerical results indicate that the new element model exhibits remarkable robustness against mesh distortion and inherent geometric adaptability, outperforming existing standard finite element and virtual element models while delivering highly competitive computational accuracy relative to hybrid virtual element models, thereby establishing itself as a reliable and efficient computational tool for continuum solid mechanics.</p>\u0000 </div>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149310878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Geometric Multigrid Preconditioner for Discontinuous Galerkin Shifted Boundary Method","authors":"Michał Wichrowski","doi":"10.1002/nme.70459","DOIUrl":"https://doi.org/10.1002/nme.70459","url":null,"abstract":"<p>I introduce a geometric multigrid preconditioner for the Shifted Boundary Method (SBM) designed to solve PDEs on complex geometries. While the SBM simplifies mesh generation by using a nonconforming background grid, it often results in nonsymmetric linear systems, a complete analysis of which is missing from the literature, especially with regard to the algebraic solution and preconditioning of high-order versions of SBM. Standard multigrid methods with pointwise smoothers prove ineffective for such systems due to the localized perturbations introduced by the shifted boundary conditions. To address this challenge, we introduce a discontinuous Galerkin (DG) formulation for SBM that enables the design of a cell-wise multiplicative smoother within an <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>h</mi>\u0000 <mi>p</mi>\u0000 </mrow>\u0000 <annotation>$$ hp $$</annotation>\u0000 </semantics></math>-multigrid framework. The element-local nature of DG methods naturally facilitates cell-wise correction, which can effectively handle the local complexities arising from the boundary treatment. Numerical results for the Poisson equation demonstrate favorable performance with mesh refinement for linear (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>p</mi>\u0000 <mo>=</mo>\u0000 <mn>1</mn>\u0000 </mrow>\u0000 <annotation>$$ p=1 $$</annotation>\u0000 </semantics></math>) and quadratic (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>p</mi>\u0000 <mo>=</mo>\u0000 <mn>2</mn>\u0000 </mrow>\u0000 <annotation>$$ p=2 $$</annotation>\u0000 </semantics></math>) elements in both 2D and 3D, with iteration counts showing mild growth. However, challenges emerge for cubic (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>p</mi>\u0000 <mo>=</mo>\u0000 <mn>3</mn>\u0000 </mrow>\u0000 <annotation>$$ p=3 $$</annotation>\u0000 </semantics></math>) elements, particularly in 3D, where the current smoother shows reduced effectiveness.</p>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/nme.70459","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149311280","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Tensor-Train Compression of a Volume Integral Eddy Current Problem in Discontinuous Galerkin","authors":"Anthony Torosyan, Bertrand Bannwarth, Olivier Chadebec, Jean-Michel Guichon, Gérard Meunier, Jean-René Poirier","doi":"10.1002/nme.70456","DOIUrl":"https://doi.org/10.1002/nme.70456","url":null,"abstract":"<p>The paper presents a low-rank tensor compression strategy based on the tensor-train (TT) format for the integral operator arising in eddy current problems. Unlike classical matrix compression techniques such as the fast multipole method (FMM) or hierarchical matrices (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>ℋ</mi>\u0000 </mrow>\u0000 <annotation>$$ mathscr{H} $$</annotation>\u0000 </semantics></math>-matrices), which typically compress only far-field interactions, the proposed approach compresses the entire operator, leading to significantly reduced memory requirements. The method relies on a tensorization step that transforms the discretized integral operator into a higher-order tensor represented in TT format. A major challenge of tensor-based compression for integral operators is the limited flexibility of tensorization, as it imposes strong constraints on the number of degrees of freedom. These constraints are particularly restrictive for conforming discretizations, where degrees of freedom are associated with mesh facets, as commonly encountered in eddy current formulations. To overcome this limitation, a discontinuous Galerkin discretization is employed, allowing the degrees of freedom to be defined at the cell level and enabling tensorization on more general meshes. The TT representation further allows the definition of an efficient matrix–vector product, which can be combined with an iterative solver to compute the solution of the resulting linear system. Numerical results for a thin plate, a torus with square cross-section, and the COMPUMAG Team 7 benchmark demonstrate a significant reduction in memory consumption compared to an FMM-based conforming discretization. While the solver performance is slightly lower, more advanced TT algebraic tools offer promising perspectives for further improvements.</p>","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/nme.70456","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149312801","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ya Zhao, Yanqi Yang, Xiaoxiang Wang, Hao Li, Qitao Tang
{"title":"Thermodynamics-Informed Neural Identification of Path-Dependent Interface Laws From Boundary-Only Observations","authors":"Ya Zhao, Yanqi Yang, Xiaoxiang Wang, Hao Li, Qitao Tang","doi":"10.1002/nme.70460","DOIUrl":"https://doi.org/10.1002/nme.70460","url":null,"abstract":"<div>\u0000 \u0000 <p>Progressive interface degradation governs the long-term load-bearing capacity of fully grouted rock anchors, yet field pullout tests yield only boundary load and displacement. Internal stress and damage fields remain experimentally inaccessible. This article presents TIINN (thermodynamics-informed interface neural network), a framework that provides a thermodynamically constrained estimation of path-dependent interface constitutive laws from boundary-only cyclic pullout observations. TIINN couples a Bregman energy network, a bounded memory damage operator (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>D</mi>\u0000 <mo>∈</mo>\u0000 <mo>[</mo>\u0000 <mn>0</mn>\u0000 <mo>,</mo>\u0000 <mn>1</mn>\u0000 <mo>)</mo>\u0000 </mrow>\u0000 <annotation>$$ Din left[0,1right) $$</annotation>\u0000 </semantics></math>), and a field distillation network with a differentiable finite-difference solver. Equilibrium is satisfied to numerical tolerance via Newton iteration, and gradients propagate to constitutive parameters through implicit differentiation, eliminating the PDE/BC loss-weight tuning inherent to penalty-based physics-informed methods. Synthetic benchmarks confirm high-fidelity constitutive recovery with boundary <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <msup>\u0000 <mrow>\u0000 <mi>R</mi>\u0000 </mrow>\u0000 <mrow>\u0000 <mn>2</mn>\u0000 </mrow>\u0000 </msup>\u0000 <mo>></mo>\u0000 <mn>0</mn>\u0000 <mo>.</mo>\u0000 <mn>9999</mn>\u0000 </mrow>\u0000 <annotation>$$ {R}^2>0.9999 $$</annotation>\u0000 </semantics></math> and damage-field correlation <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <msub>\u0000 <mrow>\u0000 <mi>ρ</mi>\u0000 </mrow>\u0000 <mrow>\u0000 <mi>D</mi>\u0000 </mrow>\u0000 </msub>\u0000 <mo>></mo>\u0000 <mn>0</mn>\u0000 <mo>.</mo>\u0000 <mn>97</mn>\u0000 </mrow>\u0000 <annotation>$$ {rho}_D>0.97 $$</annotation>\u0000 </semantics></math>. Ablation demonstrates that thermodynamic architectural constraints reduce fit variability by two orders of magnitude relative to unconstrained networks. Application to field rock-anchor cyclic pullout data improves the in-sample <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <msup>\u0000 <mrow>\u0000 <mi>R</mi>\u0000 </mrow>\u0000 <mrow>\u0000 <mn>2</mn>\u0000 </mro","PeriodicalId":13699,"journal":{"name":"International Journal for Numerical Methods in Engineering","volume":"127 21","pages":""},"PeriodicalIF":3.3,"publicationDate":"2026-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149312804","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}