0337 Nonlinear Structural Analysis with Machine Learning Based on Graph Neural Network and Structural Topology
Yuan-Tung Chou, Jimmy G. Jean, Kai-Hung Chang, Wei-Tze Chang and Chuin-Shan Chen
Nonlinear analysis allows structural engineers to understand the target force-/displacement-resistance capacity of a structure. However, it is a time-consuming process that includes not only model establishment but also computation, especially for novice engineers. Furthermore, the computation time of analysis depends on structure complexity and the degree of nonlinearity. In this study, we explore novel and rapid surrogates for structural analysis engines using machine learning methods. The Graph Neural Network (GNN) model was chosen due to its capacity of describing the structural relationship and force path between structural elements and dealing with structural responses under external loadings or prescribed displacements. Two types of graph topology, “coordinates as vertices”(CaV) and “elements as vertices” (EaV), are evaluated in terms of memory consumption, computation speed, and descriptive power. After training the model with random-generated steel structures of different numbers of stories, spans, and element lengths under lateral force or displacement, we show that the graph neural network model can achieve high engineering accuracy while reducing the computation time compared to structural analysis engines using traditional nonlinear analysis methods.
0338 Seismic Response Analysis and Shaking Table Tests of a Storage Tank Considering Fluid-Solid Coupling and Soil-Pile Interaction Effects
Liang-Yu Lu, Wei-Qing Wang, Hsuan-Teh Hu, Chun-Lin Wu, Chang-Ching Chang and Chia-Han Chen
The high-efficiency and stable power generation mode of nuclear energy is still an indispensable part of the global power supply system, and its economy and convenience are also accompanied by a high degree of risk. Taiwan and Japan are located at the same plate boundary zone, and earthquake activities are frequent. The safety of nuclear power plants when earthquakes become more important. The tank container in the nuclear reactor will be amplified by the shaking mode of the liquid inside, and it is easier to damage. In this reserch, the large-scale biaxial shear box of National Center for Research on Earthquakes Engineering (NCREE) was used to conduct shaking table tests in consideration of the fluid-solid coupling and soil structure interaction effects of the storage tank, and to observe the impact of the fluid sloshing mode of the storage tank.
The finite element method analysis software ABAQUS/Explicit is used for time-domain explicit dynamic analysis. According to the different methods of soil simulation, it is divided into physical finite element soil model and soil spring model. In addition to the fluid-solid coupling CEL analysis, the interaction effect of the soil structure is also considered, and the state of the storage tank under the seismic action during the test is truly presented. First, the significant frequency of the model is compared with the ABAQUS modal analysis solution to verify the accuracy of the model, and then a quasi-static analysis is performed. The quasi-static analysis model is input into sine waves close to the effective frequency of the fluid and the effective frequency of the sand. Analyze the soil acceleration amplification effect of the structure and fluid sloshing mode, and compare with the shaking table test results.
Keywords : ABAQUS, Soil-Pile Interaction, Fluid-Solid Coupling, Soil Spring,Shear Box Test, Shaking Table Test
0330 Multi-Degree-of-Freedom Force-Displacement Mixed Control Analysis via a Novel Use of the Zero-Length Elements
Ming-Chieh Chuang, Keh-Chyuan Tsai and Kung-Juin Wang
To mimic the complex behavior of a structure subjected to the multi-axial loadings, a simulation method for the multi-degree-of-freedom (MDOF) force-displacement mixed control analysis was proposed by the Taiwan National Center for Research on Earthquake Engineering (NCREE) researchers. Conceptually, the MDOF force-displacement mixed control analysis of a structure can be coped with the solutions for constraint problems in the finite-element method. However, it is difficult to be implemented in the existing structural analysis programs such as PISA3D and OpenSees. Thus, the NCREE researchers proposed a novel method that can offer the convenient and fast solution. In the proposed method, the substitute structure with the artificial supports is intentionally added for solving the constraint problem with the penalty function. In particular, the existing zero-length elements (e.g., 6DJoint element in PISA3D) are utilized to represent the artificial supports. Thus, the MDOF force-displacement mixed control analysis can be conducted using the existing element.
The effectiveness of the proposed method can be demonstrated via the simulation of the steel panel damper (SPD) substructure of the hybrid simulation in 2019. In the hybrid simulation, a 2.8-m-tall SPD specimen was tested using the configuration of the reaction wall and strong floor with actuators at the NCREE. The loads along MDOF were synchronously imposed on the SPD top end for producing the target deformation under seismic loading. Specifically, the SPD specimen remained zero axial (or vertical) load while it was subjected to the loads for displacement controls for the in‐plane translational and rotational degrees of freedom.
Through the proposed method, the MDOF force-displacement mixed control analysis of the SPD specimen that exerts a large plastic deformation was conducted. As shown in the figure below, the satisfactory simulation results of the histories of displacement and rotation of the SPD top end can be obtained via the proposed method with the calibrated parameters. Moreover, the proposed method could enable the MDOF force-displacement mixed control analyses using the commonly used analysis programs such as OpenSees, SAP2000 and ETABS.