Abstract: The design of struts in soft soil foundation pits is predominantly manually performed, resulting in a low level of automation. With the rapid development of computing technology, intelligent design methods have been employed to improve the efficiency of this process. To facilitate the basic implementation of automated design methods for strut layouts, an automated design approach based on graph neural networks (GNNs) was proposed. First, a graph-based representation method was established to accurately depict the beam-column topological connections. Four datasets comprising four typical strut arrangements were established. Subsequently, the GNN was trained using the aforementioned datasets. Three GNN models, including the characteristics of the connections between columns, were selected to enable link prediction of the beams between columns. Finally, the proposed method was applied to automatically generate beam structures for a secondary foundation pit planned in Shanghai. Finite element analysis was performed to validate the compatibility of the proposed methods. The results showed that the GNN-based automated design of strut layouts in foundation pits can provide rational and efficient arrangements that satisfy the regulatory design and construction requirements.

Key words: graph neural network (GNN), strut, graph representation, automated design, ?nite element analysis