ABSTRACT
The objective of a bridge design is to produce a safe bridge that is elegant and satisfies all functionality requirements, at a cost that is acceptable to the owner. Topology optimization has emerged as a method to generate innovative and high-performance structural forms, thereby offering a fresh perspective on bridge design. However, its practical application is hindered by the substantial computational costs associated with ultra-high-dimensional iterations, and it tends to prioritize the mechanical properties of the structure while neglecting aesthetic considerations. Consequently, it is difficult to meet the practical needs of engineering structural design. In light of these limitations, this study proposes a deep learning method for topology optimization design of bridge structures guided by aesthetics, with the aim of efficiently design bridge structures that exhibit both exceptional mechanical properties and visually appealing appearance. To achieve this, the study first leverages designers’ experience and intuition to evaluate the aesthetic quality of existing bridge designs, and a dataset of bridge designs labeled with their aesthetic qualities is created. Then, a neural network for evaluating the aesthetic quality of bridge structures is constructed, and the model is trained using the established dataset. Subsequently, a deep learning-based topology optimization agent model for bridge structures design is developed, so as to improve the efficiency of topology optimization. Finally, the study couples the aesthetic quality evaluation network with the topology optimization design agent model, a novel topology optimization design method for bridge pylons guided by aesthetics is established. This novel approach enables the efficient generation of structural design solutions that possess exceptional mechanical and aesthetic properties, thereby significantly elevating the level of bridge structure design and offering extensive potential for practical application.
