Axial Compressor Rotor Optimization Using a Novel Ensemble of Surrogates-Based Infill Criterion

A novel infill criterion for so-called ensemble of surrogates-based optimization is proposed and applied in practice for an aerodynamic compressor rotor design optimization. The ensemble uses a combined approach based on different radial basis functions and aims to reduce prediction errors through weighted linear combinations of radial basis functions. The update strategy uses a new hybrid custom metric termed α, which incorporates information about each surrogate’s local agreement through correlation coefficients and also information about the global accuracy of each ensemble combination through the root-mean-square error. Surrogate models are searched using a hybrid optimizer, i.e., with a genetic algorithm and sequential quadratic programming, and proposed update points are evaluated using the high-fidelity black box function. The results are compared with established optimization approaches and the best design is analyzed further in terms of the flow physics. Results show that α-based ensemble of surrogates approaches are particularly efficient for large-scale cases, where other types of surrogates such as Kriging models are onerous to construct.