Machine learning and finite element integration-driven surrogate model for fluid-structure interaction seismic response analysis of aqueduct structures

•Methodological Innovation: A synergistic TFSI modeling framework integrates multiphysics simulations and geometric parameterization, trained with 12,600 datasets.•Algorithm Advancement: The improved sand cat swarm optimization algorithm (ISCSOBP) achieves 78 % higher accuracy than traditional BP ne...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Results in engineering Jg. 27; S. 106176
Hauptverfasser: Huang, Liang, Li, Ge, Guan, Yujian, Jiao, Weili, Gong, Shengjia, Xu, Shizhan
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.09.2025
Schlagworte:
Aqueduct
Machine learning
Seismic response
Surrogate model
Two-way fluid-structure interaction
Aqueduct
Two-way fluid-structure interaction
Surrogate model
Seismic response
Machine learning
ISSN:2590-1230, 2590-1230
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•Methodological Innovation: A synergistic TFSI modeling framework integrates multiphysics simulations and geometric parameterization, trained with 12,600 datasets.•Algorithm Advancement: The improved sand cat swarm optimization algorithm (ISCSOBP) achieves 78 % higher accuracy than traditional BP neural networks, reducing computation time to 1 % of conventional FEM.•Engineering Impact: The surrogate model demonstrates high precision (maximum absolute error: 0.2 mm, relative error <3 %). The constructed integrated surrogate model has a maximum absolute error of <0.2 mm. Based on the analysis and calculations from 20 seismic waves, when the water level height-to-width ratio is <0.26, 80 % of the conditions have a seismic mitigation effect, while when it is greater than 0.52, 90 % of the conditions experience increased vibration. These findings provide a reference for the seismic design of aqueducts. The fluid-structure interaction effects in aqueduct structures under seismic excitation constitute a critical challenge in hydraulic engineering seismic analysis. While conventional numerical approaches such as the TFSI model suffer from computational inefficiency, simplified theoretical frameworks such as the Housner model fail to accurately capture the dynamic coupling mechanisms between impulsive and sloshing masses in fluid-structure systems, resulting in a persistent efficiency-accuracy trade-off in dynamic response prediction. To address this, we propose a collaborative machine learning-finite element modeling framework: First, a geometric feature parameterization method converts Boundary Surface Equation-defined aqueduct geometries into machine-interpretable inputs. Second, multiphysics-coupled FEM simulations generate 12,600 training samples. Third, a parameter-optimized machine learning architecture establishes a surrogate model for FSI-governed seismic responses. Experimental results demonstrate the surrogate model achieves 1 % computational time of conventional FEM with below 3 % dynamic prediction errors, 78.7 % higher accuracy than baseline algorithms, and enhanced numerical stability. This breakthrough provides an innovative paradigm for efficient seismic assessment of complex hydraulic structures, substantially advancing the engineering practicality of aqueduct dynamic response prediction.
ISSN:2590-1230
2590-1230
DOI:10.1016/j.rineng.2025.106176