Surrogate-enhanced simulation of aircraft in trimmed state

This paper presents a surrogate-enhanced methodology for high-fidelity simulation of aircraft in trimmed state. Surrogates of the aerodynamic loads are used to increase robustness and reduce the number of costly high-fidelity simulations necessary during trim simulations. This is accomplished in two steps: (1) by finding a suitable estimate for the solution by global and fast local searches performed on surrogates, and (2) by applying a hybrid local search to check the potential solution and to further increase its quality if necessary. For the local searches in both steps, a Newton method is employed, whereas a genetic algorithm is used only in the first step. For the Newton method, the entries of the Jacobian are provided by means of finite difference approximations and the complex-step derivative method (CSD) based on the surrogates, and it is shown how to make CSD applicable to surrogates based on radial basis functions (RBFs). In the hybrid search, the derivatives are calculated by means of surrogates, while high-fidelity simulations are used for the main calculations of the Newton method. Updating the surrogates with the results of the latter is found to significantly accelerate the convergence of the hybrid local search. The methodology is applied to challenging trim test cases, in which several of the control surfaces of a generic fighter aircraft are deflected during trim simulation. Control surface deflection is performed via RBF-based mesh deformation, and RBFs are also used for surrogate modeling, thereby fostering code modularity and re-use.