Sensitivity-based Generation of Pareto Fronts for Design of Powered Aircraft Subject to a Comprehensive Set of Loads

By virtue of using efficient methods to compute the design sensitivities, such as the coupled adjoint methods, gradient-based optimization techniques allow aircraft designers to efficiently obtain an optimum design, that satisfies all considered constraints. The more constraints and disciplines engaged, the higher the reliability of the optimization output. Decision makers in aircraft industry, however, prefer to look at Pareto fronts rather than one optimum design before making their critical decisions, since such diagrams provide better understanding of the main trade-offs and compromises between the different targets that the aircraft is supposed to satisfy. Gradient-free algorithms have better reputation for generating Pareto fronts because these fronts result naturally. These algorithms are, however, relatively inefficient, even for single disciplinary high-fidelity optimizations. They are restricted by the number of design parameters and the size of the design space. Engaging more disciplines, more design parameters and higher fidelity computational models, to increase the reliability of the outputs, can only increase the computational cost of gradient-free algorithms, and quickly announce them to be not usable for the generation of Pareto fronts. This study investigates the generation of Pareto fronts efficiently, using gradient-based algorithms on the industry-relevant aircraft; AIRBUS XRF-1. The approach starts several optimizations with different weighting of objectives, in parallel. The multidisciplinary design chain, which is used in this study, was connected by experts in aerodynamics, structure and loads, propulsion and overall aircraft design. The generation of Pareto front was completed successfully for a multi-point multidisciplinary optimization task, where the use of gradient-free algorithms would have been computationally infeasible. As expected and mentioned in the literature, some regions of the Pareto front were not satisfactorily covered by employing this approach, which shows the necessity to look into different gradient-based approaches that search for optimal designs along the Pareto front.