A New Approach for Deviation Modeling in Compressors: Sensitivity-Correlated Principal Component Analysis

Abstract

Studies on the geometry variation-related compressor uncertainty quantification (UQ) have often used dimension reduction methods, such as the principal component analysis (PCA), for the modeling of deviations. However, in the PCA method, the main eigenmodes were determined based only on the statistical behavior of geometry variations. While this process can cause some missing modes with a small eigenvalue, it is much more sensitive to blade aerodynamic performances, and thereby reducing the reliability of the UQ analysis. Hence, a novel geometry variation modeling method, named sensitivity-correlated principal component analysis (SCPCA), has been proposed. In addition, by means of the blade sensitivity analysis, the weighting factors for each eigenmode were determined and then used to modify the process of the PCA. As a result, by considering the covariance of geometry variations and the performance sensitivity, the main eigenmodes could be determined and used to reconstruct the blade samples in the UQ analysis. With 98 profile samples measured at the midspan of a high-pressure compressor rotor blade, both the PCA and SCPCA methods were employed for the UQ analysis. The results showed that, compared to the PCA method, the SCPCA method provided a more accurate reconstruction of sensitive deviations, leading to an 11.8% improvement in evaluating the scatter of the positive incidence range, while also maintaining the accuracy of the uncertainty assessment for other performances.

Keywords

• geometric deviation
• uncertainty quantification
• principal component analysis
• assessing accuracy
• compressor aerodynamics