TY - GEN
T1 - GNN-Based Structural Dynamics Simulation for Modular Buildings
AU - Zhang, Jun
AU - Zhang, Tong
AU - Wang, Ying
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022.
PY - 2022
Y1 - 2022
N2 - Modular buildings are made up of standardized building sections manufactured in a controlled environment. Their advantages include speedy construction process, cost-effectiveness, and higher quality. Further, it provides an opportunity to perform data-driven numerical simulations through a standardized process. Due to its capability of topological generalization, a Graph Neural Network (GNN) based approach is proposed in this study as an innovative structural dynamics simulation tool. The proposed approach can predict the dynamic response of structures with different topologies by changing the relationship matrix. To demonstrate its effectiveness, three spring-mass systems, with 3-DOF, 6-DOF, and 10-DOF, are used as examples of modular buildings. The dynamic response data of the 3-DOF system are used to train a GNN model. After necessary topological adaptation, the model is then used to predict the dynamic response of the 6-DOF system and the 10-DOF system, without any new training process. The results show that the proposed approach can predict the dynamic responses of structures with different topologies with a very low peak mean square error (PMSE), which is less than 0.01. It has the potential to enhance the generalization capabilities of data-driven numerical simulation methods.
AB - Modular buildings are made up of standardized building sections manufactured in a controlled environment. Their advantages include speedy construction process, cost-effectiveness, and higher quality. Further, it provides an opportunity to perform data-driven numerical simulations through a standardized process. Due to its capability of topological generalization, a Graph Neural Network (GNN) based approach is proposed in this study as an innovative structural dynamics simulation tool. The proposed approach can predict the dynamic response of structures with different topologies by changing the relationship matrix. To demonstrate its effectiveness, three spring-mass systems, with 3-DOF, 6-DOF, and 10-DOF, are used as examples of modular buildings. The dynamic response data of the 3-DOF system are used to train a GNN model. After necessary topological adaptation, the model is then used to predict the dynamic response of the 6-DOF system and the 10-DOF system, without any new training process. The results show that the proposed approach can predict the dynamic responses of structures with different topologies with a very low peak mean square error (PMSE), which is less than 0.01. It has the potential to enhance the generalization capabilities of data-driven numerical simulation methods.
KW - AI for Science
KW - Graph neural networks
KW - Modular construction
KW - Structural dynamic response prediction
KW - Structural health monitoring
UR - https://www.scopus.com/pages/publications/85142791863
U2 - 10.1007/978-3-031-18913-5_19
DO - 10.1007/978-3-031-18913-5_19
M3 - 会议稿件
AN - SCOPUS:85142791863
SN - 9783031189128
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 245
EP - 258
BT - Pattern Recognition and Computer Vision - 5th Chinese Conference, PRCV 2022, Proceedings
A2 - Yu, Shiqi
A2 - Zhang, Jianguo
A2 - Zhang, Zhaoxiang
A2 - Tan, Tieniu
A2 - Yuen, Pong C.
A2 - Guo, Yike
A2 - Han, Junwei
A2 - Lai, Jianhuang
PB - Springer Science and Business Media Deutschland GmbH
T2 - 5th Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2022
Y2 - 4 November 2022 through 7 November 2022
ER -