Surrogate Modelling for Wing Planform Multidisciplinary Optimisation Using Model-Based Engineering

Abstract

Optimisation is aimed at enhancing aircraft design by identifying the most promising wing planforms at the early stage while discarding the least performing ones. Multiple disciplines must be taken into account when assessing new wing planforms, and a model-based framework is proposed as a way to include mass estimation and longitudinal stability alongside aerodynamics. Optimisation is performed with a particle swarm optimiser, statistical methods are exploited for mass estimation, and the vortex lattice method (VLM) with empirical corrections for transonic flow provides aerodynamic performance. Three surrogates of the aerodynamic model are investigated. The first one is based on radial basis function (RBF) interpolation, and it relies on a precomputed database to evaluate the performance of new wing planforms. The second one is based on an artificial neural network, and it needs precomputed data for a training step. The third one is a hybrid model which switches automatically between VLM and RBF, and it does not need any precomputation. Its switching criterion is defined in an objective way to avoid any arbitrariness. The investigation is reported for a test case based on the common research model (CRM). Reference results are produced with the aerodynamic model based on VLM for two- and three-objective optimisations. Results from all surrogate models for the same benchmark optimisation are compared so that their benefits and limitations are both highlighted. A discussion on specific parameters, such as number of samples for example, is given for each surrogate. Overall, a model-based implementation with a hybrid model is proposed as a compromise between versatility and an arbitrary level of accuracy for wing early-stage design.