Feature-Mapping Topology Optimization with Neural Heaviside Signed Distance Functions

Topology optimization plays a crucial role in designing efficient and manufacturable structures. Traditional methods often yield free-form voids that, although providing design flexibility, introduce significant manufacturing challenges and require extensive post-processing. Conversely, feature-mapping topology optimization reduces post-processing efforts by constructing topologies using predefined geometric features. Nevertheless, existing approaches are significantly constrained by the limited set of geometric features available, the variety of parameters that each type of geometric feature can possess, and the necessity of employing differentiable signed distance functions. In this paper, we present a novel method that combines Neural Heaviside Signed Distance Functions (Heaviside SDFs) with structured latent shape representations to generate manufacturable voids directly within the optimization framework. Our architecture incorporates encoder and decoder networks to effectively approximate the Heaviside function and facilitate optimization within a unified latent space, thus addressing the feature diversity limitations of current feature-mapping techniques. Experimental results validate the effectiveness of our approach in balancing structural compliance, offering a new pathway to CAD-integrated design with minimal human intervention.

Designing physical structures — like parts for machines — requires making choices that balance strength, efficiency, and how easily something can be manufactured. Specialized software aids design decisions but still requires manual input to produce manufacturable designs.One promising approach, called Topology Optimization, uses mathematical techniques to decide where material should go in a design to make it strong and lightweight. However, the shapes it produces are often too complex to be manufactured.Our work explores how machine learning can help close this gap. We are developing a framework that emulates real engineers' design principles by creating parts whose voids are represented as combinations of manufacturable geometric building blocks. Specifically, we use a kind of AI model known as a Variational Autoencoder to learn how to represent basic geometric building blocks — like triangles and quadrangles — in a common format. This makes it easier to combine and modify them in ways that meet engineering and manufacturing needs.Rethinking shape representation, our method enables efficient, manufacturable designs. We release open-source code to foster further work and drive automated tools that cut engineer time and accelerate product development.