Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations
Developing fast and accurate computational models to simulate intricate physical phenomena has been a persistent research challenge. Recent studies have demonstrated remarkable capabilities in predicting various physical outcomes through machine learning-assisted approaches. However, it remains chal...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing LLC
2023-12-01
|
| Series: | APL Machine Learning |
| Online Access: | http://dx.doi.org/10.1063/5.0167014 |
| _version_ | 1797367037165568000 |
|---|---|
| author | Zeqing Jin Bowen Zheng Changgon Kim Grace X. Gu |
| author_facet | Zeqing Jin Bowen Zheng Changgon Kim Grace X. Gu |
| author_sort | Zeqing Jin |
| collection | DOAJ |
| description | Developing fast and accurate computational models to simulate intricate physical phenomena has been a persistent research challenge. Recent studies have demonstrated remarkable capabilities in predicting various physical outcomes through machine learning-assisted approaches. However, it remains challenging to generalize current methods, usually crafted for a specific problem, to other more complex or broader scenarios. To address this challenge, we developed graph neural network (GNN) models with enhanced generalizability derived from the distinct GNN architecture and neural operator techniques. As a proof of concept, we employ our GNN models to predict finite element (FE) simulation results for three-dimensional solid mechanics problems with varying boundary conditions. Results show that our GNN model achieves accurate and robust performance in predicting the stress and deformation profiles of structures compared with FE simulations. Furthermore, the neural operator embedded GNN approach enables learning and predicting various solid mechanics problems in a generalizable fashion, making it a promising approach for surrogate modeling. |
| first_indexed | 2024-03-08T17:12:10Z |
| format | Article |
| id | doaj.art-bce72142a4d34e239c3eef361f141894 |
| institution | Directory Open Access Journal |
| issn | 2770-9019 |
| language | English |
| last_indexed | 2024-03-08T17:12:10Z |
| publishDate | 2023-12-01 |
| publisher | AIP Publishing LLC |
| record_format | Article |
| series | APL Machine Learning |
| spelling | doaj.art-bce72142a4d34e239c3eef361f1418942024-01-03T19:54:29ZengAIP Publishing LLCAPL Machine Learning2770-90192023-12-0114046109046109-710.1063/5.0167014Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulationsZeqing Jin0Bowen Zheng1Changgon Kim2Grace X. Gu3Department of Mechanical Engineering, University of California, Berkeley, California 94720, USADepartment of Mechanical Engineering, University of California, Berkeley, California 94720, USAHyundai Motor Company, 150 Hyundaiyeonguso-ro, Namyang, Hwaseoung, Gyeonggi 18280, South KoreaDepartment of Mechanical Engineering, University of California, Berkeley, California 94720, USADeveloping fast and accurate computational models to simulate intricate physical phenomena has been a persistent research challenge. Recent studies have demonstrated remarkable capabilities in predicting various physical outcomes through machine learning-assisted approaches. However, it remains challenging to generalize current methods, usually crafted for a specific problem, to other more complex or broader scenarios. To address this challenge, we developed graph neural network (GNN) models with enhanced generalizability derived from the distinct GNN architecture and neural operator techniques. As a proof of concept, we employ our GNN models to predict finite element (FE) simulation results for three-dimensional solid mechanics problems with varying boundary conditions. Results show that our GNN model achieves accurate and robust performance in predicting the stress and deformation profiles of structures compared with FE simulations. Furthermore, the neural operator embedded GNN approach enables learning and predicting various solid mechanics problems in a generalizable fashion, making it a promising approach for surrogate modeling.http://dx.doi.org/10.1063/5.0167014 |
| spellingShingle | Zeqing Jin Bowen Zheng Changgon Kim Grace X. Gu Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations APL Machine Learning |
| title | Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations |
| title_full | Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations |
| title_fullStr | Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations |
| title_full_unstemmed | Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations |
| title_short | Leveraging graph neural networks and neural operator techniques for high-fidelity mesh-based physics simulations |
| title_sort | leveraging graph neural networks and neural operator techniques for high fidelity mesh based physics simulations |
| url | http://dx.doi.org/10.1063/5.0167014 |
| work_keys_str_mv | AT zeqingjin leveraginggraphneuralnetworksandneuraloperatortechniquesforhighfidelitymeshbasedphysicssimulations AT bowenzheng leveraginggraphneuralnetworksandneuraloperatortechniquesforhighfidelitymeshbasedphysicssimulations AT changgonkim leveraginggraphneuralnetworksandneuraloperatortechniquesforhighfidelitymeshbasedphysicssimulations AT gracexgu leveraginggraphneuralnetworksandneuraloperatortechniquesforhighfidelitymeshbasedphysicssimulations |