Integration of High-Fidelity Analysis Tools in MDO Frameworks for HPC

Kurzfassung

Two framework-based MDO approaches currently being developed and implemented at DLR are presented in this paper. Both target the integration of high-fidelity simulation methods into multidisciplinary aircraft design optimization processes. Owing to the different levels of optimization problem complexity, the levels of code integration differ significantly. In the first case, the multiphysics HPC environment FlowSimulator and its high-fidelity analysis plugins are systematically modularized to facilitate the flexible construction of high-fidelity MDO processes using OpenMDAO's Modular Analysis and Unified Derivatives (MAUD) approach. Based on a FlowSimulator-OpenMDAO synthesis we rebuild and run an existing gradient-based steady-state aeroelastic shape optimization process that takes advantage of the FlowSimulator environment to perform the multidisciplinary evaluations in the optimization process. The major HPC components involved, which are plugins to the FlowSimulator environment, are the viscous CFD method TAU, a finite-element structural model, a linear elasticity-based mesh deformation method, an algorithm enforcing trimmed flight equilibria, methods for surface mesh interpolation, and a reduced-order model (ROM) of a parametric CAD description that is used in the loop. The second MDO approach presented here relies on a new framework solution also focusing on the HPC integration of high-fidelity simulation and optimization methods. The approach has been conceived to handle strongly collaboration-intensive MDO for transport aircraft design. A specific MDO problem representation augmented by a highly parallel process assembly and execution protocol (called Cybermatrix MDO) is defined. This MDO framework is currently being implemented and evaluated on a HPC cluster in conjunction with three disciplinary sub-processes within the DLR project VicToria. These are the above-mentioned gradient-based high-fidelity aeroelastic optimization process governing the aerodynamic wing shape, the optimization process that is in charge of the structural sizing of lifting surfaces, and a process determining the design loads. Based on the two multidisciplinary optimization approaches, with the Cybermatrix MDO framework integrating the FlowSimulator-enabled MDO approach, the two strategies are being analyzed and discussed in order to carve out the potentials and limitations of the two MDO solutions for HPC.