An MDAO Framework Approach Extended to Unsteady Multiphysics: Key Ingredients and Examples

Abstract

State-of-the-art MDAO frameworks like \textup{OpenMDAO} and \textup{GEMSEO} facilitate the setup and the gradient-enabled solution/optimization of large multidisciplinary problems, which are predominantely of steady-state type. Time-dependent MDAO problems, however, pose additional challenges, particularly in conjunction with cost-intensive high-fidelity simulations. This work demonstrates how advanced MDAO-framework capabilities can be transferred from the steady-state to the unsteady domain. To this end, an MDAO framework approach is presented integrating \textup{OpenMDAO} with the HPC ecosystem \textup{FlowSimulator} including high-fidelity simulation plugins like the CFD-software \textup{CODA}. Among the key capabilities of the time-accurate framework approach are a complete system representation on the framework level resolving all input- and output dependencies to yield a monolithic time-resolved view, and checkpointing-enabled reverse gradient accumulation over time. A gradient-based optimization was conducted in the framework for an elementary Van der Pol oscillator case carrying out an unsteady reverse-in-time sensitivity analysis. Moreover, it was successfully shown and verified that the framework is capable of performing unsteady adjoint viscous/turbulent CFD computations for a NACA0012 and the CRM/DPW-5 wing-body configuration, paving the way for large adjoint-enabled unsteady MDAO applications.