Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks
This paper presents a physics-informed training framework for projection-based Reduced-Order Models (ROMs). We extend the original PROM-ANN architecture by complementing snapshot-based training with a FEM-based, discrete physics-informed residual loss, bridging the gap between traditional projection...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-05-01
|
| Series: | Axioms |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-1680/14/5/385 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849711737616465920 |
|---|---|
| author | Nicolas Sibuet Sebastian Ares de Parga Jose Raul Bravo Riccardo Rossi |
| author_facet | Nicolas Sibuet Sebastian Ares de Parga Jose Raul Bravo Riccardo Rossi |
| author_sort | Nicolas Sibuet |
| collection | DOAJ |
| description | This paper presents a physics-informed training framework for projection-based Reduced-Order Models (ROMs). We extend the original PROM-ANN architecture by complementing snapshot-based training with a FEM-based, discrete physics-informed residual loss, bridging the gap between traditional projection-based ROMs and physics-informed neural networks (PINNs). Unlike conventional PINNs that rely on analytical PDEs, our approach leverages FEM residuals to guide the learning of the ROM approximation manifold. Our key contributions include the following: (1) a parameter-agnostic, discrete residual loss applicable to nonlinear problems, (2) an architectural modification to PROM-ANN improving accuracy for fast-decaying singular values, and (3) an empirical study on the proposed physics-informed training process for ROMs. The method is demonstrated on a nonlinear hyperelasticity problem, simulating a rubber cantilever under multi-axial loads. The main accomplishment in regards to the proposed residual-based loss is its applicability on nonlinear problems by interfacing with FEM software while maintaining reasonable training times. The modified PROM-ANN outperforms POD by orders of magnitude in snapshot reconstruction accuracy, while the original formulation is not able to learn a proper mapping for this use case. Finally, the application of physics-informed training in ANN-PROM modestly narrows the gap between data reconstruction and ROM accuracy; however, it highlights the untapped potential of the proposed residual-driven optimization for future ROM development. This work underscores the critical role of FEM residuals in ROM construction and calls for further exploration on architectures beyond PROM-ANN. |
| format | Article |
| id | doaj-art-ece2cec5741e493b9bd1dea8a53502f2 |
| institution | DOAJ |
| issn | 2075-1680 |
| language | English |
| publishDate | 2025-05-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Axioms |
| spelling | doaj-art-ece2cec5741e493b9bd1dea8a53502f22025-08-20T03:14:32ZengMDPI AGAxioms2075-16802025-05-0114538510.3390/axioms14050385Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural NetworksNicolas Sibuet0Sebastian Ares de Parga1Jose Raul Bravo2Riccardo Rossi3Department of Civil and Environmental Engineering (DECA), Universitat Politècnica de Catalunya, 08034 Barcelona, SpainDepartment of Civil and Environmental Engineering (DECA), Universitat Politècnica de Catalunya, 08034 Barcelona, SpainDepartment of Civil and Environmental Engineering (DECA), Universitat Politècnica de Catalunya, 08034 Barcelona, SpainDepartment of Civil and Environmental Engineering (DECA), Universitat Politècnica de Catalunya, 08034 Barcelona, SpainThis paper presents a physics-informed training framework for projection-based Reduced-Order Models (ROMs). We extend the original PROM-ANN architecture by complementing snapshot-based training with a FEM-based, discrete physics-informed residual loss, bridging the gap between traditional projection-based ROMs and physics-informed neural networks (PINNs). Unlike conventional PINNs that rely on analytical PDEs, our approach leverages FEM residuals to guide the learning of the ROM approximation manifold. Our key contributions include the following: (1) a parameter-agnostic, discrete residual loss applicable to nonlinear problems, (2) an architectural modification to PROM-ANN improving accuracy for fast-decaying singular values, and (3) an empirical study on the proposed physics-informed training process for ROMs. The method is demonstrated on a nonlinear hyperelasticity problem, simulating a rubber cantilever under multi-axial loads. The main accomplishment in regards to the proposed residual-based loss is its applicability on nonlinear problems by interfacing with FEM software while maintaining reasonable training times. The modified PROM-ANN outperforms POD by orders of magnitude in snapshot reconstruction accuracy, while the original formulation is not able to learn a proper mapping for this use case. Finally, the application of physics-informed training in ANN-PROM modestly narrows the gap between data reconstruction and ROM accuracy; however, it highlights the untapped potential of the proposed residual-driven optimization for future ROM development. This work underscores the critical role of FEM residuals in ROM construction and calls for further exploration on architectures beyond PROM-ANN.https://www.mdpi.com/2075-1680/14/5/385reduced-order model (ROM)physics-informed neural networks (PINNs)artificial neural network (ANN)projection-based model reductionproper orthogonal decomposition (POD) |
| spellingShingle | Nicolas Sibuet Sebastian Ares de Parga Jose Raul Bravo Riccardo Rossi Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks Axioms reduced-order model (ROM) physics-informed neural networks (PINNs) artificial neural network (ANN) projection-based model reduction proper orthogonal decomposition (POD) |
| title | Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks |
| title_full | Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks |
| title_fullStr | Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks |
| title_full_unstemmed | Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks |
| title_short | Discrete Physics-Informed Training for Projection-Based Reduced-Order Models with Neural Networks |
| title_sort | discrete physics informed training for projection based reduced order models with neural networks |
| topic | reduced-order model (ROM) physics-informed neural networks (PINNs) artificial neural network (ANN) projection-based model reduction proper orthogonal decomposition (POD) |
| url | https://www.mdpi.com/2075-1680/14/5/385 |
| work_keys_str_mv | AT nicolassibuet discretephysicsinformedtrainingforprojectionbasedreducedordermodelswithneuralnetworks AT sebastianaresdeparga discretephysicsinformedtrainingforprojectionbasedreducedordermodelswithneuralnetworks AT joseraulbravo discretephysicsinformedtrainingforprojectionbasedreducedordermodelswithneuralnetworks AT riccardorossi discretephysicsinformedtrainingforprojectionbasedreducedordermodelswithneuralnetworks |