Multi-fidelity design optimisation of a transonic compressor rotor

摘要

This paper demonstrates the application of Kriging based approximations to the aerodynamic shape optimisation of a transonic compressor rotor where multiple fidelities of analysis are available. An extension to the co-Kriging method is implemented, and attention is paid to the efficient construction of the approximate models. The initial data sample within each level of the model is selected using space-filling designs, and the covariance structure of the co-Kriging models is obtained using a maximum likelihood estimation based on gradient optimisation with adjoint functions. An efficient leave-one-out cross-validation method is used to estimate model accuracy and is used to guide initial model refinement.