Parallel Adjoint Framework for Aerodynamic Shape Optimization of Component-Based Geometry

We present a parallel adjoint framework for aerodynamic shape optimization problems using an embedded-boundary Cartesian mesh method. The design goals for the framework focus on an efficient and systematic integration of the underlying software modules. By linearizing the geometric constructors used in intersecting triangulated components, we develop a robust approach for computing surface shape sensitivities of complex configurations. The framework uses multilevel parallelism in sensitivity computations. Serial and parallel codes are executed concurrently to speedup gradient computations. A variety of optimizers are supported, and geometric modelers can be either CAD-based or CAD-free. We present design examples involving sonic-boom mitigation and nacelle integration for transport aircraft involving over 160 design variables and using up to 256 processors. This is an important step in our research toward making aerodynamic shape optimization tools available to the broader aerodynamics community instead of CFD specialists only.