Development of helium turbine loss model based on knowledge transfer with Neural Network and its application on aerodynamic design

Abstract

Helium turbines are widely used in the Closed Brayton Cycle for power generation and aerospace applications. The primary concerns of designing highly loaded helium turbines include choosing between conventional and contra-rotating designs and the guidelines for selecting design parameters. A loss model serving as an evaluation means is the key to addressing this issue. Due to the property disparities between helium and air, turbines utilizing either as working fluid experience distinct loss mechanisms. Consequently, directly applying gas turbine experience to the design of helium turbines leads to inherent inaccuracies. A helium turbine loss model is developed by combining knowledge transfer and the Neural Network method to accurately predict performance at design and off-design points. By utilizing the loss model, design parameter selection guidelines for helium turbines are obtained. A comparative analysis is conducted of conventional and contra-rotating helium turbine designs. Results show that the prediction errors of the loss model are below 0.5% at over 90% of test samples, surpassing the accuracy achieved by the gas turbine loss model. Design parameter selection guidelines for helium turbines differ significantly from those based on gas turbine experience. The contra-rotating helium turbine design exhibits advantages in size, weight, and aerodynamic performance.