Original scientific paper
Graph Neural Network for Stress Predictions in Stiffened Panels
Yuecheng Cai
; The University of British Columbia, 6250 Applied Science Ln #2054, Vancouver, BC V6T 1Z4
*
Jasmin Jelovica
orcid.org/0000-0002-8396-941X
; The University of British Columbia, 6250 Applied Science Ln #2054, Vancouver, BC V6T 1Z4
* Corresponding author.
Abstract
Graph neural network (GNN) is a particular type of neural network which processes data that can be
represented as graphs. This allows for efficient representation of complex geometries that can change
during conceptual design of a structure or a product, such as ship structures, replacing computationally
expensive finite element analysis (FEA) in optimization. In this study, we demonstrate how GNN
can be used to predict stress distributions in stiffened panels with varying geometries under patch
loading, for which we use Graph Sampling and Aggregation (GraphSAGE) network. Parametric study
is performed to examine the effect of structural geometry on the prediction performance. Our results
demonstrate the immense potential of graph neural networks with the proposed graph embedding
method as robust reduced-order models for 3D structures.
Keywords
Machine learning, Deep learning, Graph neural networks, Stiffened panels, Structural analysis
Hrčak ID:
317231
URI
Publication date:
16.5.2024.
Visits: 281 *