A CAD-enabled MDAO Framework Approach for Gradient-based Aerodynamic Shape Optimization

Abstract

We aim at a seamless CAD-integration into a framework-based approach for multidisciplinary design analysis and optimization (MDAO) that allows an automated forward and reverse accumulation of AD-based gradients throughout complex high-fidelity workflows. The suggested framework approach relies on the FlowSimulator HPC ecosystem, in which a number of high-fidelity simulation tools, called plug-ins, share large mesh-based coupling data sets in memory via the FlowSimulator Data Manager (FSDM) in workflows that are MPI-parallel from end-to-end. Based on the FlowSimulator infrastructure and a granular plug-in integration, the MDAO framework OpenMDAO [1] was used to drive the CAD-enabled, gradient-based optimization in conjunction with the CFD software CODA. A fully-resolved system representation is generated for the MDAO problem at hand based on a systematic registration of plug-in input/output dependencies. In this work, the algorithmically differentiated CAD kernel OpenCascade Technology (OCCT) was integrated into the FlowSimulator ecosystem to centrally provide a gradient-enabled CAD link for all the simulation plug-ins involved in the problem. The individual FSDM surface mesh objects -- that can be (MPI) domain-decomposed -- are linked to the corresponding CAD surfaces using a reliable, meta-data enabled mesh-to-CAD association to robustly deal with fine meshes in the context of high Reynolds-number flows. Selected aerodynamic wing configurations in 2D and 3D are considered in this study to verify and demonstrate the AD-enabled sensitivity analysis with the focus on the framework integration of the CAD kernel OCCT in the context of shape optimization.