Prediction of centrifugal compressor performance curves based on multi-task Gaussian process

Abstract

The performance curves are crucial to the safety operation of the compressor. As a method to obtain the performance curves of centrifugal compressor, Computational Fluid Dynamics(CFD)has the advantages of high data accuracy, but its calculation cost is expensive. In order to reduce the calculation cost, surrogate model has been widely used to predict the performance curve of the centrifugal compressor in recent years. In previous studies, PR and EFF at various speed lines were predicted separately neglecting the similarities across speed lines. Considering the correlation between different speeds, the multi-task Gaussian Process (MTGP) method was adopted in this paper to predict the performances curves of a centrifugal compressor in a more efficient manner than the conventional models by leveraging the cross-line information.To validate the effectiveness of the method, the predictions from MTGP are compared with that of multi-fidelity method CoKriging and single-fidelity Kriging. Finally, we found that MTGP has the highest prediction accuracy under the same data cost.