Adjoint methods for aerodynamic shape optimization have been widely available for the Reynolds-averaged Navier-Stokes (RANS) equations since the late 1990s. Arguably, they are the most efficient methods for the calculation of high-dimensional gradient vectors when only a small number of quantities of interest are involved. Within university research groups, such methods have been used for sensitivity analysis in steady and unsteady RANS settings, for aero-structural and aero-acoustics problems, and even for aero-thermal-structural problems including conjugate heat transfer. Yet industry has not embraced the use of these methods in its day-to-day activities and processes. In this talk we provide a historical perspective of the development of adjoint methods for many industrial applications including transonic commercial aircraft design, supersonic low-boom design, turbomachinery, and aero-thermal-structural reliability problems in nozzles of advanced aircraft. We then focus on the key remaining research issues that must be tackled to make adjoint use more pervasive: adjoint solution robustness, consistency of sensitivity information in multi-disciplinary settings, the cost-effective treatment of unsteady problems, and the efficient mapping on HPC architectures. A future state is envisioned where adjoint solutions are completely automatic and transparent and are provided to the user the requests them regardless of parameterization and the quantities of interest to be differentiated.
展开▼
