Abstract

We propose a comprehensive process chain that considerably reduces the effort and time to successfully interpret topology optimization results and convert them into CAD design proposals. The process chain extracts a wireframe model from a topology optimization result based on a voxelization approach which represents a unique strategy in this field and allows the use of efficient image processing algorithms in the subsequent process steps. The process chain furthermore suggests a suitable cross section for each beam based on the topological geometry as well as stress results calculated during the optimization run. As a first step in the extraction of a wireframe model, the topological finite element mesh-based model is filtered using the relative element density results from the topology optimization, deleting all elements with a relative density below a configurable threshold. The three-dimensional design space of the model is then discretized with a regularly spaced grid – voxels. The filtered FE mesh is converted into a voxel-based representation, thus gaining semi-independence from the arbitrary representation of the FE mesh whose elements can vary in size, form and connectivity. Currently, the FE mesh can consist of any combination of two-dimensional tria and quad elements as well as three-dimensional tetrahedral elements. The now voxel-based representation of the topological model is then thinned using a well-established skeletonization algorithm. From the still voxel-based but one-voxel-thick skeleton lines, structural voxel areas are detected where three or more skeleton lines join together using flood filling algorithms. The centers of these voxel structures are then addressed and connected with lines, creating the wireframe model. With the wireframe model, it is possible to identify beams in the topological model. Each beam in the topology optimization model is cut and the resulting cross section geometrically analyzed to get a first approximate cross section area and shape. This step highly depends on the chosen element density threshold. Furthermore, the topology optimization generates element stresses for each finite element. This data is also analyzed and compared to a defined set of example stress distributions (e.g. tension-compression or bending). Based on the approximate cross section area and the stress analysis results, a suitable cross section for each beam is automatically chosen from a database (e.g. I-section, round, square, …). Future work includes refining the cross section analysis, possibly considering machine learning algorithms, and further processing of the wireframe model, e.g. further FE analyses and a final design proposal in a CAD format.