A Spanwise Loss Model of Turbine Cascade with Tip Clearance Based on Machine Learning

Yucheng Chen,Donghai Jin,Xingmin Gui
DOI: https://doi.org/10.1007/978-981-97-3998-1_90
2024-01-01
Abstract:The tip clearance leakage flow of turbine cascade significantly affects the spanwise distribution of circumferential averaged total pressure loss at its outlet. Therefore, this paper aims to build a spanwise loss model, using a combination of machine learning methods, to effectively improve the simulation accuracy of through-flow calculation of S2 stream surface. Based on the blade profile of the first stage of an aeroengine at 90% span, using variables such as tip clearance height and camber angle, the flow field is calculated by numerical simulation, a database is constructed, including nearly 150 samples. Feature engineering is applied to select 8 variables such as prediction of AMDC model of tip clearance leakage loss, and 3 intermediate variables such as pressure loss coefficient of tip leakage flow and secondary flow. Then the random forest algorithm is used to screen to input variables in prediction models. And BP neural network is used as estimator, with genetic algorithm optimizing its initial weight and threshold. Finally, a spanwise loss distribution model with 10 estimators and a selector is constructed. The results show that the model has good regression and prediction ability for spanwise loss distribution curve, with R-2 greater than 0.98 and MAPE less than 2.5%. Thanks to the completeness of parameter coverage in the database, the combination of machine learning methods, and knowledge related to tip leakage flow, the model in this paper has good generalization ability. It can provide effective technical support for the development of turbine through-flow-design and simulation analysis methods.
What problem does this paper attempt to address?