Aero-structural discrete adjoint sensitivities in su2 using algorithmic differentiation

In recent years, extensive research has been conducted on the integration of high-fidelity aero-structural analysis models in Multidisciplinary Design Analysis and Optimization (MDAO) applications for aircraft design. In this paper, we outline a number of key requirements for the implementation of such methods in industrial-scale design processes. These requirements have emerged from efforts to couple SU2 with external structural solvers, such as NASTRAN and the Airbus structural MDAO suite Lagrange, for gradient-based optimization using large-scale MDAO frameworks. We also detail the implementation and validation of a new discrete adjoint flow solver in SU2, providing previously unavailable partial derivatives required for the calculation of coupled aero-structural gradients. As part of this work, the Python interface of SU2 was revised and extended to enable easy in-memory integration of its primal and adjoint mesh deformation and flow solvers. The new discrete adjoint solver is demonstrated in sensitivity analyses for multiple test cases, varying in scale, geometric complexity, and flow conditions. Excellent agreement is observed between the new residual-based formulation and the existing fixed-point approach, as well as with finite-difference approximations. The newly available partial derivatives are exact, efficient to evaluate, and not susceptible to truncation and cancellation error.