Performance Study for High-Fidelity Static Aeroelastic Analyses Using the OpenMDAO Framework and the FlowSimulator HPC Ecosystem

In this paper, we evaluate the multidisciplinary solver performance of an MDAO framework approach integrating the FlowSimulator HPC simulation ecosystem, the CFD Software by ONERA, DLR, Airbus (CODA) and OpenMDAO for large-scale aerodynamic and static aeroelastic analyses. For a robust and efficient implicit solution of stiff multidisciplinary problems, nested solver stacks are applied and evaluated with a tight/monolithic coupling between the disciplines. Two numerical studies are presented in this paper. First, preconditioned Newton-Krylov methods implemented in the MDAO framework approach are applied to solve the RANS equations for the DPW-5/Common Research Model (CRM) wing-body configuration on a sequence of hybrid meshes L2-L5, resulting in 61 million to 1.2 billion degrees of freedom handled by the MDAO framework. The performance of the framework-based solver integrating the CFD software CODA as a single implicit component is compared against the corresponding solution algorithms directly implemented in the performance-optimized CFD solver CODA. In a second MDA solver study, we consider a simplified version of the aeroelastic NASA R4/SDT engine test case in combination with a CSM finite-element (FE) solid model of the flexible rotor blade in b2000++ and a fully-turbulent RANS model in CODA implicitly coupled in the MDAO framework environment. We investigate both algorithmic and parallel performances for multidisciplinary, Newton-GMRES solvers in combination with both parallel linear block-Jacobi and sequential linear block-Gauss-Seidel preconditioning. The parallel efficiency of the concurrent linear block-Jacobi preconditioner outperformed the algorithmic advantage of the linear block-Gauss-Seidel preconditioner for this case.