Intelligent beam layout design for frame structure based on graph neural networks

The layout design of the frame structure beams is a critical task in frame structure design. Traditional automatic layout methods often rely on established rules. However, the predefined rules are often incomplete, and the conflicts and priorities between different constraints are often unclear. Consequently, it is difficult for traditional automatic methods to meet the challenges of flexible layout of structures with free planar shapes. The beam–column connection of the frame structures exhibits the topological characteristics of graphs. A graph neural network is a data-driven geometric deep learning algorithm that is suitable for addressing non-Euclidean data such as graphs, thus providing a new solution for the beam layout design of frame structures. Therefore, this study proposes an intelligent plan layout design method for frame beams based on a graph neural network. A large-scale dataset of the frame structure layout was considered for the neural network training. Graph representation methods for frame structures are discussed, and a novel graph neural network model for beam layout design is proposed. The test results show that the proposed beam layout design method has high accuracy, and case studies of real-world frame structures show that the outcome of the proposed method is comparable to engineer's design.