Publications

We present an efficient and robust strategy to control discretization error during aerodynamic shape optimization (ASO) using a high-order discretization with curved mesh adaptation. During aerodynamic shape optimization, it is important to have an accurate solution to prevent discretization error from polluting the optimum. High-order methods are promising because they offer increased accuracy for a given mesh. Mesh adaptation further improves the efficiency of high-order methods. These high-order methods require curved meshes to properly capture the simulated geometry and a mesh adaptation process that can generate curved meshes. Adapting these curved meshes needs to be robust as any failures will require human intervention inside the automated optimization loop. In this work, we introduce two key advances. The first advancement is a method for robustly splitting elements on geometric boundaries, which is necessary to properly capture the geometry. The second advancement is a concurrent mesh adaptation strategy that automatically balances accuracy and computational efficiency throughout the optimization process. Results for transonic airfoil optimization demonstrate that our method reduces the number of expensive fine-mesh evaluations by 75-90% compared to traditional approaches.