Session: 36-04 UQ & Sensitivity Analysis - Part 1

Submission Number: 176734

Global Sensitivity Analysis on Compressor Performance Through Advanced ANCOVA Under Consideration of Correlated Geometric Parameters 

Variations in compressor airfoil geometry, arising from manufacturing scatter and erosion, are a major source of uncertainty in aerodynamic performance. To better understand and quantify how such input uncertainties affect system behaviour, global sensitivity analysis methods are employed. These methods decompose output variance into contributions from individual inputs and their interactions, making them essential tools in engineering for identifying influential parameters. However, when inputs are correlated, widely used approaches such as Sobol indices and Shapley values provide limited insight into how these correlations contribute to variance.

To address this, the advanced ANCOVA method is introduced, which attributes variance caused by correlations directly to the responsible parameter pairs, enabling a clearer interpretation of correlated effects.

To evaluate and compare the performance of the proposed method with established approaches, three aerodynamic cases are considered. The first involves a 2.5-stage compressor, where the second rotor blade row geometry is varied through correlated profile parameters. Monte Carlo simulations are carried out using extended Latin hypercube sampling, followed by a sensitivity analysis of the isentropic efficiency using the aforementioned measures. In addition, the effect of the number of deterministic model runs on the accuracy of the sensitivity measures is examined.

The second aerodynamic case examines a 10-stage high-pressure compressor. The impact of varying the size of all rotor tip gaps on the isentropic efficiency is investigated. Correlations between tip gap clearances are varied to highlight the superior interpretability of the advanced ANCOVA method compared to other established SA methods when inputs are correlated.

The third aerodynamic case considers variations in the profile parameters of all rotor geometries in the 10-stage high-pressure compressor. This high-dimensional setting involves 140 correlated parameters, with the isentropic efficiency and total pressure ratio being analysed using the aforementioned sensitivity measures. The results demonstrate that the advanced ANCOVA method outperforms established approaches by decomposing the significance of each input into structural and correlation-driven contributions.

Presenting Author: Roman Kalbitz Technical University of Dresden

Presenting Author Biography: 2018 - 2024: Diploma in Aerospace Engineering, Technical University of Dresden

Feb 2024 - Aug 2024: Diploma Thesis at MTU Aero Engines - Design study of radial compressors as the final stage in aircraft engines

Since Oct 2024: Research Associate at the Chair of Turbomachinery and Flight Propulsion, Technical University of Dresden – focusing on the development and application of probabilistic methods to analyze geometric variability in compressor blades

Authors:

Roman Kalbitz Technical University of Dresden
Max Dittmann MTU Aero Engines AG
Matthias Voigt Technical University of Dresden
Ronald Mailach Technical University of Dresden