Design Optimization of Truss Structures Using Artificial Neural Networks
Date
2023-09-22
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
One of the primary objectives of structural design optimization is to achieve a design possessing the lowest possible weight, while it can safely withstand the effects of external loads. In the case of a truss of a specific topology, the role of an optimization algorithm is to determine the configuration and number of the truss elements as well as their cross-sectional areas. In this study, a novel model is proposed, by which the main optimization problem is decomposed into two more manageable problems: a size optimization within a shape optimization problem. A Deep Neural Network (DNN) is trained to approximate the optimal cross-sectional areas of the elements of a truss with a given shape and support positions. Furthermore, truss structures are characterized by pin joints connected by truss members, a concept that can be analogized to vertices and edges in a mathematical graph. Leveraging this analogy, a Graph Neural Network (GNN) is utilized to exploit the advantages of representing trusses as graphs. Specifically, a graph neural network-based surrogate model integrated with Particle Swarm Optimization (PSO) algorithm is developed to approximate nodal displacements of trusses during the design optimization process. Several truss examples are used to investigate the validity and effectiveness of the proposed optimization techniques in comparison with conventional FEM-based models.
Description
Keywords
design optimization, artificial neural network, graph neural network, deep neural network, genetic algorithm, particle swarm optimization, shape optimization, size optimization, surrogate model, truss structures
Citation
Nourian, N. (2023). Design optimization of truss structures using artificial neural networks (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.