Intelligent design for component size generation in reinforced concrete frame structures using heterogeneous graph neural networks

Abstract

Traditional reinforced concrete (RC) frame design depends on extensive engineering experience and iterative verification processes, often resulting in significant inefficiencies. The diversity in the topologies and behaviors of structural components further presents considerable obstacles to effective machine learning applications in design. This paper introduces an approach using heterogeneous graph neural networks (HetGNNs) to automate and optimize the dimensioning of frame components. This method captures the distinct frame topologies by developing a precisely tailored heterogeneous graph node representation. Leveraging a unique dataset derived from engineering drawings, the HetGNN model learns to size the component sections accurately. It is demonstrated that this method offers a transformative improvement in the efficiency, accuracy, and cost-effectiveness of structural design while adhering to design standards. The size design of RC frame structures can be completed in under one second, with an average size deviation of around 50 mm (one module) compared to those designed by engineers.