A method of local surrogate models is proposed to improve the accuracy of the prediction of aerodynamic fields for parametric systems with nonlinear behaviors. The so-called Local Decomposition Method makes use of Machine Learning tools to group the solutions of the training sample into homogeneous clusters and to separate the parameter space in function of the shape of the solution. A local surrogate model is built on each subdomain, providing accurate predictions for each characteristic regime. Moreover, a specific resampling strategy has been developed to focus on the more challenging clusters. The Local Decomposition Method has been applied on the AS28G aircraft configuration and shows promising results in term of flow regime detection and global accuracy.
展开▼
