A spanwise loss model for axial compressor stator based on machine learning

Abstract

In the early stage of aircraft engine design, the through-flow method is an important tool for designers. The accuracy of the through-flow method depends heavily on the accuracy of the loss model. However, most existing models cannot (or cannot well) provide the spanwise loss distribution. To construct an effective spanwise loss model, both turbomachinery knowledge and machine learning skills were used in this paper. A large number of numerical simulations were carried out to build a database containing more than 1000 compressor cascade numerical samples. Secondary flow intensity was introduced as the independent variable to carry out feature engineering. A model containing a selector based on support vector machine regression and estimators based on K-nearest neighbor regression was constructed. Numerical test set and design data of two former high-pressure core compressors were used for validation. Results suggest that the spanwise loss model show good consistency with both numerical test set and data of two former compressors. It can reflect the influence of secondary flow and can also predict both value and trend of total pressure loss coefficient well, with mean absolute error general around or less than 1% and R2 (coefficient of determination) more than 0.8 on the test set. Especially when dealing with loss coefficient at mid-span position, the model shows even better performance, with R2 over 0.97 on the test set. And the selector of the model can well classify the samples, predict the intensity of secondary flow and help estimators to capture the phenomenon that end-wall secondary flow extends to the mid-span.