State-of-the-art review of automated structural design optimization

Abstract

<p>Automated and intelligent structural design optimization has been a heated research topic in structural engineering community. The automated structural design may learn from existing design drawings and finite element models of previous design, infuse design experts鈥?knowledge with design standards and codes, and notably reduce the time and difficulty of structural design compared to conventional approach. This study provides a review of current automated structural design optimization approaches, which may be categorized as the approaches based on finite element model and the deep-learning approach based on human design dataset. For the design optimization based on finite element model, the gradient-based algorithm and gradient-free algorithm are illustrated and compared. The selection of objective function and constraint functions for structural design optimization are summarized. The parallel computing method developed based on high-performance computing resources are also summarized. In addition, the deep learning approaches, which directly generate preliminary structural design drawings based son architectural drawings and datasets of human design results are also summarized. Major architectures in this research field are discussed, including generative adversarial networks (GAN), diffusion models and Variational Auto-Encoder (VAE). The combination between deep learning approach and conventional finite-element model-based approach are also discussed and the future development trends and potential challenges are discussed.</p>