Abstract

There are many challenges today in obtaining better quality outcomes in generative design using conventional tetrahedron meshes. In addition, adding more elements to obtain better shapes typically means slower performance and when running jobs on the cloud and this translates to higher costs. Also, tetrahedron meshes are not ideal for certain manufacturing constraints such as milling and extrusion. One solution is to use a voxelized mesh which approximately conforms to the input design space envelop using voxel elements. One benefit of this is the ability to use fast interpolation from a finer design space mesh to a coarser one used for FEA, resulting in a better outcome both in aesthetics and the objective function without a loss in accuracy. Our voxelized mesh concept results in a more accurate application of boundary conditions which allows using a smaller number of voxel elements to perform accurate FEA computations. The coarser FEA mesh also results in better overall performance and reduced cloud costs. The finer optimization design space results in smoother shapes especially with complex starting geometry resulting more detail yielding a more optimal design. Using a voxelized mesh also allows for faster optimization operations associated with manufacturing constraints and more compliant milling and extrude manufacturing constraints. A side benefit of using interpolation is improved multidisciplinary optimization where other FEA physics such normal modes, buckling, heat transfer, and CFD can be coupled for fully compliant design. In this presentation, we combine linear statics, normal modes, and buckling to design various structures. The linear static analysis handles the stress and displacement constrains and runs in parallel with a normal modes and buckling solution all as separate processes communicating using Inter Process Communication or IPC. The current architecture assumes that the design problem has a single objective function which is minimize compliance with multiple design and manufacturing constraints using different FEA physics. The main topology optimization engine is controlled by a fast independent GOCM optimization processor, while constraints are handled in multiple dependent FEA processes which calculate their own adjoint sensitivities and pass data to the independent process using IPC. Optimal designs are further achieved by using an adjustable volume fraction constraint which adjusts up or down to satisfy all design constraints.