Research on Uncertainty Analysis and Robust Optimization Design Method of Aero-Engine Compressors

Abstract

The compressor is a critical component of an aero-engine, with its performance directly influencing the overall engine efficiency and stability. The inherent uncertainties in the production and development processes of compressors often lead to deviations in average performance from the design specifications and significant performance variability. Therefore, it is of great significance to improve the comprehensive performance of aero engines to establish an uncertainty model for the compressor, carry out uncertainty quantification and sensitivity analysis, identify the influencing laws and mechanisms of key parameters, and then reduce performance sensitivity and control performance dispersion through aerodynamic robustness optimization. Taking the high-load compression system as the object, this thesis carries out geometric uncertainty modeling based on the measured processing data of compressor blades, explores the impact of uncertainties on aerodynamic performance, and constructs an uncertainty analysis and robust optimization framework applicable to different dimensions. The main research work is as follows:

(1) Firstly, this thesis carries out research on geometric uncertainty modeling of processing deviation of compressor blades. Aiming at the problem of scarcity of measured geometric data of the compressor and the unrealistic uncertainty model, a geometric feature recognition and statistical method is developed, and the measured data of the compressor blade is analyzed and statistically analyzed to establish a more realistic geometric uncertainty model. The results show that there are significant uncertainty deviations in processing and manufacturing, but they generally conform to the Gaussian distribution.

(2) Subsequently, this thesis carries out research on uncertainty quantification and sensitivity analysis of geometry and aerodynamics of compressors. Aiming at the problem that the effects and mechanisms of uncertainty factors in compressors are still unclear, a general and complete integrated uncertainty analysis method of uncertainty modeling–quantification–sensitivity analysis–mechanism research is constructed by combining uncertainty quantification, comprehensive sensitivity analysis and data mining. The research discovers the significant impact of uncertainty on compressor aerodynamic performance and reveals the nonlinear and strongly coupled influence mechanism of uncertain factors on aerodynamic performance.

(3) Then, this thesis carries out research on high-dimensional multi-fidelity uncertainty quantification methods for compressors. Aiming at the problem of large sample size and high computational cost required for high-dimensional uncertainty quantification research of compressors, a high-fidelity surrogate model and a dimensionality-reduced low-fidelity surrogate model coupling bidirectional search strategy and feature selection method are established respectively. The two are combined through model management strategy to form a dimensionality-reduced multi-fidelity surrogate model, which significantly reduces the required sample size while maintaining high model prediction accuracy. This method is applied to compressors, revealing the mechanism of high-dimensional uncertainty leading to deterioration of compressor average performance, successfully reducing the uncertainty dimension and the number of sampling points by one order of magnitude, and realizing the quantification of high-dimensional uncertainty under multiple working conditions.

(4) Finally, this thesis carries out research on the robust optimization design method of compressors. Aiming at the problem that the robust optimization process and parameters are complex, the calculation is large, and the optimization direction is black box, the dynamic machine learning model, the model interpretation method and the trust region global optimization algorithm are combined to establish an efficient and high-precision global robust optimization method with invisible parameters and interpretable optimization process, which significantly reduces the complexity and time of the aerodynamic robustness optimization process. By combining different surrogate models with optimization algorithms, an efficient robust optimization framework suitable for different dimensions is established. Starting from the best universality commonly used optimization frameworks gradually transition to multi-fidelity optimization frameworks that are more suitable for high-dimensional practical problems, the compressor robust optimization design system is enriched.

Date of Award	23 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xingen LU (External Supervisor) & Penghao DUAN (Supervisor)