Derivation of Pressure Loss Models for Turbine Center Frames via an L1-Regularized Regression

Marian Staggl, Wolfgang Sanz, Peter Leitl, Maximillian Kurzthaler, Paul Pieringer

Research output: Chapter in Book/Report/Conference proceedingConference paperpeer-review

Abstract

The turbine center frame (TCF) is a key component of modern aero engines, guiding the airflow from the high-pressure to the low-pressure turbine. Due to an increasing flow area, it has a diffusing effect, making it prone to flow separations. Open literature offers various layout guidelines based on performance correlations, but in order to generate parsimonious models, many authors restrict their investigations to a small parameter space while neglecting other important factors. On the other hand, a correlation based on a large number of variables might be more accurate yet very impractical. An L1-regularized least squares fit (LASSO) yields a possibility to balance the contradictory demands for a simple and accurate model. In the current work, the LASSO method is used to correlate the total pressure loss of a straight strutted TCF with various geometrical and inlet-flow parameters. The strength of the regularization allows setting the tradeoff between the model's complexity and accuracy. In this way crucial parameters for the performance of the TCF could be identified and a simple polynomial pressure loss model for the design optimization could be found.
Original languageEnglish
Title of host publicationProceedings of the 15th European Turbomachinery Conference ETC15, Budapest, Hungary. ETC2023-118
Publication statusPublished - Jun 2023
Event15th European Turbomachinery Conference: ETC15 - Budapest, Hungary
Duration: 24 Apr 202328 Apr 2023

Conference

Conference15th European Turbomachinery Conference
Abbreviated titleETC15
Country/TerritoryHungary
CityBudapest
Period24/04/2328/04/23

Fingerprint

Dive into the research topics of 'Derivation of Pressure Loss Models for Turbine Center Frames via an L1-Regularized Regression'. Together they form a unique fingerprint.

Cite this