In-depth Generative Design Combining Generative Models with Topology Optimization

Abstract

Recently, deep learning has been used in a number of design optimization study domains. The 
need and efficacy of using deep learning in the field of generative design (or design exploration) 
research are discussed in this article. This paper suggests a deep generative design framework 
based on artificial intelligence (AI) that may provide a wide range of aesthetically pleasing and 
engineering performance-optimized design solutions. In order to investigate novel design 
possibilities, the suggested framework iteratively combines topology optimization with 
generative models (such as generative adversarial networks, or GANs), producing a huge number 
of designs from a little amount of prior design data. Anomaly detection may also assess the 
originality of created designs, assisting designers in selecting from a variety of design 
possibilities. The suggested methodology is validated using the 2D wheel design issue as a case 
study. Compared to earlier generative design techniques, the framework produces designs that 
are more resilient, diverse, and aesthetically pleasing.