Session: 13-13-01: Functional Origami and Kirigami-inspired Structures and Materials

Paper Number: 166840

Topology Optimization Using Graph Neural Network in Additive Manufacturing Constraint 

Topology optimization (TO) has emerged as a powerful tool for designing high-performance structures by optimizing material distribution to satisfy specific performance criteria. While traditional topology optimization methods
are widely applied in industries such as aerospace, automotive, and architecture, their high computational costs pose
significant challenges, especially in large-scale problems with complex constraints. These limitations are particularly
critical in additive manufacturing (AM), where the design of support structures plays a crucial role in minimizing
deformation, reducing material waste, and enhancing production efficiency. Overhanging structures provide a significant issue for additive manufacturing. Supports are commonly utilized to alleviate this condition, providing numerous
benefits. They decrease distortion in overhangs and secure parts to the build platform. Recent advancements in deep
learning offers a pathway to overcome these challenges by significantly improving computational efficiency without
sacrificing accuracy.
This paper focuses on a novel Graph Neural Network (GNN)-based topology optimization framework specifically
designed for the creation of support structures in additive manufacturing. By representing the design domain as a
graph, where nodes correspond to material elements and edges capture their spatial relationships, the GNN effectively
models the complex geometries of support structures. A Fourier projection layer is incorporated to enhance the
resolution of fine structural details, ensuring the generation of precise and efficient support designs. The optimization
process is guided by a hybrid loss function that combines compliance minimization with volume constraints, achieving
an optimal balance between material efficiency and structural performance.
A key feature of the proposed framework is the seamless integration of finite element analysis (FEA) within the
GNN architecture, allowing for direct compliance calculation and efficient sensitivity analysis through automatic differentiation. Unlike traditional topology optimization methods, which require manual sensitivity analysis, the GNN-based approach automatically computes these sensitivities using automatic differentiation provided by frameworks
like PyTorch. This research also demonstrates that employing a fully connected neural network instead of a GNN
framework leads to slower convergence. The suggested algorithm’s usefulness is demonstrated using numerical examples. Numerical results demonstrate the effectiveness of the method in generating optimized support structures that
minimize deformation under applied forces, reduce material usage, and shorten the design-to-manufacture pipeline.
By focusing on the critical task of support structure design in AM, this framework offers a scalable, computationally efficient, and practical solution for advancing the capabilities of topology optimization in modern manufacturing
processes.
Keywords: Topology optimization, Deep learning, Graph Neural network, Fully connected Neural Network, Sensitivity Analysis, Support Structure, Additive Manufacturing
 

Presenting Author: Saquib Ahmad Bhuiyan University of North Carolina at Charlotte

Presenting Author Biography: Ph.D. student at the Mechanical Engineering department at the University of North Carolina at Charlotte.

Authors:

Saquib Ahmad Bhuiyan University of North Carolina at Charlotte
Alireza Tabarraei University of North Carolina at Charlotte