Geometric Feature Identification from Topology Optimization Results

1. Abstract Topology optimization yields an overall layout of a structure in the form of discrete density (e.g., SIMP) or continuous boundary geometry (e.g., level-set method). One of important drawbacks, however, is that it leads to a geometry with zigzag boundaries and/or irregular shapes, which is difficult to be interpreted for manufacturability, as well as to be utilized in subsequent applications such as shape optimization. It is considered the most significant bottleneck to interpret topology optimization results and to produce a parametric CAD model that can be used for shape optimization. The objective of this paper is to interpret geometric features out of a topology design to minimize human intervention in producing a parametric CAD model. The active contour method is first used to extract boundary segments from the greyscale image of topology optimization. Using the information of roundness and curvature of segments, simple geometric features, such as lines, arcs, circles, fillets, extrusion and sweep, are then identified. An optimization method is used to find parameters of these geometric features by minimizing errors between the boundary of geometric features and that of actual segments. Lastly, using the parametric CAD model, surrogate-based shape optimization is employed to determine the optimal shape. The entire process is automated with MATLAB and Python scripts in Abaqus, while manual intervention is needed only when defining geometric constraints and design parameters. 2-D beam and plate structures are presented to demonstrate effectiveness of the proposed methods. 2.