A Scalable MDAO Framework Approach Enabling Nested Solution Algorithms for Static Aeroelastic Coupling Scenarios

Abstract

A modular framework approach is presented that enables the use of scalable solution algorithms for high-fidelity multidisciplinary (MD) problems. It relies on the simulation plugins provided by the FlowSimulator HPC ecosystem including the CFD software CODA and the general-purpose CSM solver b2000++, among others. The MDAO framework OpenMDAO was integrated into the FlowSimulator infrastructure and a full system view was built for different static aeroelastic coupling scenarios. Based on the fine-grained integration of simulation components with access to component-wise directional algorithmic derivatives, deep solver stacks can be realized in a flexible way: on the MD level, pseudo-time relaxed Newton algorithms are combined with Krylov solvers which, in turn, are preconditioned by linear block-Gauss--Seidel schemes. In the disciplinary blocks, the (linear) solution algorithms provided by the individual plugins are used, which can be highly specialized algorithms like in the case of CFD solvers. The MDAO framework approach is MPI-parallel from end to end and allows for a user-friendly association of HPC resources with (sub)groups of the MD system. The performance of nested complex solution methods and their dependency on key solver parameters are investigated and discussed for selected 3D test cases like the LANN wing and the NASA SDT engine configuration with flexible blades in fully-turbulent flow conditions. Convergence, solution efficiency and robustness of deep MD solver stacks are systematically compared with straightforward MD solution methods such as nonlinear block-Gauss--Seidel methods: the more advanced MD algorithms are observed to be particularly favorable in conjunction with increased MD coupling strengths, indicating that they can retain algorithmic robustness and efficiency for future MDAO problems of larger scale.