Abstract: This paper presents a framework for automating shape design by combining geometric reasoning with numerical optimization. Unlike traditional parameter-based methods that require pre-defined variables, the proposed approach utilizes the medial axis transform, or "skeleton," as a flexible geometric abstraction. By treating rays extending from this skeleton to the boundary as design variables, the system can evolve shapes through significant topological changes, such as the emergence of holes or the elimination of unnecessary features, while maintaining geometric integrity.
Introduction
Shape design represents a significant challenge in engineering because the final topology of an optimal structure is often unknown at the outset. Traditional approaches in 1991 typically fell into three categories: rigid parameter-based methods, computationally expensive grid-based methods, and hard-to-generalize knowledge-based systems. This paper introduces a methodology that allows a shape to "grow" or "shrink" its own features dynamically during optimization.
Methodology: The Skeleton Abstraction
The core of the proposed system is the use of the Skeleton (Medial Axis Transform) as the primary geometric representation. The skeleton acts as an "inner spine" of a shape, consisting of all points equidistant from at least two points on the boundary.
Design variables are defined as the lengths of "rays" emanating from specific points on the skeleton to the boundary. This representation ensures the physical integrity of the shape; because rays are bounded by the skeleton, the boundary cannot self-intersect or create invalid "negative" space. The optimization process follows an iterative loop:
- Compute the current shape's skeleton.
- Define rays from the skeleton as optimization variables.
- Apply Non-Linear Programming to adjust ray lengths to maximize an objective function (e.g., stiffness).
- Update the shape and repeat.
Topological Evolution
A key advantage of this skeleton-based approach is its ability to handle topological changes naturally. As the optimization progresses, if two boundaries are pushed together until the thickness reaches zero, a protrusion can disappear. Conversely, if the boundary evolves to wrap around an exterior branch of the skeleton, a hole is formed. This allows for the synthesis of complex structures, such as hollow tubes, without explicit instructions to create a cavity.
Results: I-Beam Synthesis
The effectiveness of the method was demonstrated through several structural benchmarks. In the "Wide Flange" evolution, the system began with a simple rectangular block.
As shown in the evolution sequence, the optimization algorithm automatically identifies and removes unnecessary material while thickening the "flanges." Over several iterations, the initial block evolves into a shape closely resembling a standard structural I-beam, significantly increasing its moment of inertia (from I = 2.67 to I = 22.7).
Conclusion
By marrying geometric reasoning with numerical optimization, this work provides a robust path for automated shape synthesis. The skeleton-based representation offers a unique combination of flexibility and safety, allowing for topological discovery while preventing geometric invalidity. While computationally demanding for its time, the approach laid foundational principles for modern generative design and topology optimization.
Summary updated