Three-dimensional optimization of a 1.5-stage axial compressor based on a novel local adaptive ensemble surrogate model

Abstract

Accurately predicting the performance of a compressor is of utmost importance when utilizing intelligent optimization algorithms. To improve the prediction accuracy, a novel local adaptive ensemble surrogate model (LAESM) is proposed. In this model, independent individual surrogate model screening and weight calculation are carried out for each point to be predicted according to a unique local performance index, with the purpose of giving full play to the local advantages of different individual surrogate models. 24 numerical functions are used to test the LAESM and some other surrogate models, and it is observed that the LAESM demonstrated better accuracy and stability when compared to other surrogate models. Meanwhile, a simulation failure processing method based on SVM classification model (FP-SVM) is proposed, and a one-dimensional function is used to show the feasibility of this method. Combining the LAESM and FP-SVM, a 1.5-stage axial compressor is optimized. A total of 18 design variables and 6 objective functions are considered in the optimization, and 626 samples are calculated using the RANS method for the training. The results show that after optimization, the efficiency, pressure ratio, and stable operating range of the axial compressor are improved. By observing the flow field, it is found that the flow loss inside the compressor is obviously reduced as a result of adjusting the rotor blade profile. The method proposed in this study has the potential to serve as a reference for optimization problems in the field of turbomachinery.