Machine learning powered sketch aided design via topology optimization

Abstract

Structural topology optimization is an important design tool in the conceptual design phase of a product. However, the current topology optimization design is mostly driven strictly based on mathematical and mechanical models. Although the innovative design can be automated, it mostly lacks effective manual experience guidance. To improve the efficiency of computer-aided structural design models, this work proposes a sketch-guided topology optimization approach based on machine learning. Using neural network-based style transfer techniques, computer-digitized/hand-drawn sketches are explicitly incorporated into topology optimization in the form of constraint functions. The obtained optimization results not only satisfy the requirements of optimal mechanical properties, but also fully demonstrate the design intention and requirements of designers. Numerical examples show that the proposed approach can effectively compensate for the lack of manual experience guidance for topology optimization.