Genetic Algorithm Based Collaborative Optimization of a Tiltrotor Configuration
Stanley Orr, Prabhat Hajela · 46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference · 2005
The rotorcraft optimization problem is characterized by challenges stemming from computationally demanding analyses, complex multidisciplinary interactions, and a design space that may contain several local optima. These problems are compounded further by the additional complexity of the tiltrotor configuration. At the very outset, the aerodynamic and structural design of the proprotor blade is tightly coupled, exhibiting strong multidisciplinary interactions. The rotor and wing systems also exhibit interactions, ranging from moderate to strong. A rotor and wing design problem has been formulated that is sufficiently complex so as to represent industrial practice, and is easily scalable from moderate to large size. The details of the problem formulation have been presented in a companion paper. The current paper explores the solution of the optimal design problem. A collaborative optimization scheme was considered to handle the potentially large-scale problem and manage the multi-disciplinary design interactions. Separation of the rotor and wing into subsystems was shown to be a reasonable choice for decomposition, though approximations were necessary to ensure a manageable set of shared variables between the two sub-systems. Optimal search was based on a collaborative optimization approach that computed the sub-problem optimal design solution at each iteration, or function evaluation of the master problem. While this approach magnified the number of function evaluations, it allowed a global search of the design space when coupled to genetic algorithm optimal search at the master and sub-problem levels. Neural network function surrogates were used for the sub-problems; this provided a relatively quick run time for the sub-problem optimal searches. The strategy proposed in the paper allowed use of surrogate models in a manner that mitigated the influence of inaccuracy in their predictive capability. The strategy was shown to be relatively efficient in terms of the required number of exact function evaluations, particularly when compared to an all-in-one approach.