Uncertainty quantification based on active subspace dimensionality-reduction method for high-dimensional geometric deviations of compressors

Compressors are inevitably exposed to diverse geometric deviations from manufacturing errors and in-service degradation. Consequently, the evaluation of performance uncertainties becomes of utmost importance for compressors in engineering application. However, the presence of high-dimensional and strongly nonlinear geometric deviations poses significant challenges in efficiently and accurately assessing the performance uncertainties of compressors. This study proposes an active subspace-based dimensionality-reduction method for high-dimensional uncertainty quantification (UQ) of compressors. Based on the active subspace (AS) method, a dimensionality-reduction high-precision artificial neural network is raised to solve the dimension disaster problem for high-dimensional UQ. Additionally, a data-driven approach is used to calculate the gradient of the quantity of interest, addressing the issue of high computational cost during the AS dimensionality reduction process. Furthermore, the Shapley method is applied to explore the influence mechanism of geometric uncertainties on performance deviations of compressors. The UQ of one transonic compressor stage at design point and near stall point is conducted by the proposed method. The findings show that the original 24-dimensional uncertainties are reduced to three-dimensional uncertainties by using this method. Consequently, the required sample size is reduced by 75% while maintaining almost unchanged model accuracy. The findings reveal that the sweep and stagger deviation of the rotor are key uncertainties on the performance of the compressor. The dispersion in efficiency is attributed to variations in shock wave position and intensity, while the dispersion in total pressure ratio is primarily affected by changes in rotor work capacity. Moreover, the dispersion at near stall is 50% higher than that at design point. Therefore, when studying UQ, it is important to pay closer attention to the performance dispersion at near stall conditions.