A data-driven non-intrusive polynomial chaos for performance impact of high subsonic compressor cascades with stagger angle and profile errors

The impact of geometric variations due to manufacturing errors on the aerodynamic performance of compressor blades is considerable in engineering practice. Accurate uncertainty quantification (UQ) of aerodynamic performance based on actual statistical information of manufacturing errors is helpful for error detection, aerodynamic shape design, etc. However, the actual manufacturing errors may not be independent of each other, and their marginal distributions are often not standard. To apply UQ to dependent input variables with arbitrary distributions, the present paper first introduced a decorrelation algorithm and proposed a novel data-driven Non-Intrusive Polynomial Chaos (DNIPC) model. The subsequent tests based on high-dimensional function experiments and UQ of the performance for a high subsonic compressor cascade have validated its generality, effectiveness, and efficiency. Then, the simultaneous impact of stagger angle and profile errors on the performance of the compressor cascade was investigated by the proposed DNIPC and Sobol's sensitivity analysis. The distributions of manufacturing error were determined by kernel density estimation based on finite measurement data, then using the parametric B-spline and Bezier curves to build the manufactured geometry of the cascade. Finally, the influence of the manufacturing errors on aerodynamic performance was further discussed through the UQ of the cascade flow field. The results show that the performance impact in the off-design conditions, especially for those below the design incidence, is the most important. It suggested that the uncertainty of profile error should be checked carefully in the process of detecting machining error when the operation condition is at or below the design incidence; in contrast, the influence of stagger angle error calls for more attention when the cascade works at high incidences. The fluctuation of the overall aerodynamic loss is mainly associated with the manufacturing error near the leading edge, for it affects the velocity spike and the development of its downstream boundary layer. Therefore, the leading edge error should be the top focus.