Efficient Methods for High Fidelity Topology Optimization

This thesis presents work related to effcient numerical solution schemes and implementations of large-scale, or high-resolution, topology optimization. The purpose of pursuing effcient methods is to increase the resolution of computed
structures, which in turn enable new potential applications. The thesis has then been split into two, each part focusing on different types of computing hardware. The first part treats the implementation of topology optimization on high performance compute clusters. We therefore deal with distributed memory computing, primarily on distributed unstructured meshes. We begin by describing the multigrid preconditioner, which is a key component in solving the elasticity equations on the desired scale. Afterwards, a new filter for ensuring manufacturability by milling or casting is presented, specifically designed with distributed unstructured meshes in mind. Finally, a short case study of a direct drive wind turbine is performed. The second part treats the implementation of structural optimization on a single desktop machine. We study GPU acceleration, and how far we can push the performance of a single high-end machine by limiting the problem to structured grids while in a shared memory setting. We then move on to shape optimization of shell structures, which emphasizes how high-resolution is not necessarily required to provide meaningful structural optimization results.
Finally, we summarize the results of this work with a critical reflection on the value of high-resolution topology optimization compared to the cost of required hardware and effort of implementation.