Development of surrogate-based multidisciplinary optimization methodology with flutter constraints for aircraft conceptual design

Abstract

Aircraft concepts with high aspect ratio wings have been investigated more extensively in the recent past, as such configurations would reduce the induced drag significantly. At the same time, the high aspect ratio challenges the wing’s aero-structural stability (flutter). Therefore there is a need to perform aero-structural analysis already within the conceptual design phase, preferably in an automated fashion to enable the execution of multidisciplinary design optimization (MDO). This study has investigated methods to directly incorporate flutter speed predictions as constraints in an automated optimization process for conceptual design of aircraft. The computationally expensive flutter calculations have been modelled using different surrogate models (SM) to reduce calculation time in the automated design process. The combination of different surrogate and design of experiments (DOE) methods have been investigated and compared. Each combination of SM and DOE provides an indication of the trade-off between computation time and precision. The MultiFit software tool, developed by the Netherlands Aerospace Centre, was used to set up the different SMs. Artificial neural networks have shown the best capabilities in representing the behaviour of the flutter speed. Furthermore, surrogate-based optimization (SBO) methods have been able to find the same optimum design point at lower computational cost compared to the classical (non-surrogate) multidisciplinary feasible (MDF) MDO architecture. In an attempt to reduce optimization time even further, an adaptive surrogate modelling approach has been incorporated in the surrogate-based optimization method. The proposed method starts with a small DOE to create an initial SM. Then, an SBO is performed on this model. The value of the resulting optimum is recalculated using the complete multidisciplinary analysis and this new design point is then added to the DOE. A new surrogate model is created and an SBO is performed using the earlier found optimum as a starting point. This process is repeated until the process converges. The case study in this research has shown that adaptive surrogate-based optimization can be an efficient method to find the optimum of the objective function at a low computational cost, although more advanced enrichment and convergence criteria are recommended to improve the algorithm.