15th European Conference on Turbomachinery Fluid dynamics & Thermodynamics

Paper ID:


Main Topic:

Axial Turbines



Marian Staggl  - Graz University of Technology, Austria
Wolfgang Sanz - Graz University of Technology, Austria


A turbine center frame is a key component of modern aeroengines. It guides the airflow from the high-pressure turbine to the low-pressure turbine. Due to the increasing area along the flow path, the turbine center frame has a diffusing effect, making it prone to flow separation. The main metric to evaluate the performance of turbine center frames is the total pressure loss which is influenced by inflow conditions and its geometry. While some parameters have a significant impact, the influence of others might be negligible. The sensitivity of the pressure loss to the various parameters is of great interest for the layout and optimization of this component and is expressed by the derivatives with respect to the different parameters. If the available data on pressure loss is scattered in the parameter space, a correct estimation of the derivatives is difficult. In this work, a multivariate polynomial is defined over the various parameters to describe the pressure loss. The closed form of the polynomial allows the calculation of continuous derivatives over the whole parameter space. The polynomial is fit to the dataset via the so-called LASSO algorithm which uses a L1 regularized least squares minimization. The regularization stabilizes the model fit, dampens oscillations of the polynomials and prevents overfitting. Another remarkable property of the L1 minimization is the tendency to identify sparse models for the given data. Most of the polynomial coefficients are set to 0 by the least squares fit and drop out of the equation. The degree of sparsity can be controlled by a parameter which allows setting the tradeoff between the complexity of the polynomial and its accuracy. This can help to identify the crucial parameters and thus leading to a comprehensible and also interpretable algebraic formulation of the derivatives. A comprehensive data set of CFD simulations on turbine center frames is used to show the suitability of the method for estimating the pressure loss derivatives. The simulations cover a wide range of the given parameter space, describing geometrical variations as well as changes in the boundary conditions. It is also worth mentioning that this method of derivative estimation is not restricted to simulation results and can be applied equally well to experimental data.


Download it! Paper is available for download