Key points are not available for this paper at this time.
Abstract The aerodynamic design and performance improvement of compressors in gas turbine development, especially in multistage axial flow compressors with numerous design parameters, pose challenges relying on manual design experience. To address computational inefficiency in aerodynamic optimization of multistage turbomachinery, an aerodynamic optimization platform is developed based on the Direct Free-Form Deformation (DFFD) method. This method, offering full-geometry three-dimensional parameterization with arbitrary degrees of freedom, conducts direct manipulation of control vertex displacement, generating smooth shapes, reducing control variables, and achieving dimensionality reduction. The optimization framework combines the DFFD method with a Pre-screen data-driven single-objective optimization algorithm using differential evolution and a hybrid surrogate model. Introducing machine learning techniques significantly reduces Computational Fluid Dynamics (CFD) simulations, enhancing iteration efficiency. Applied to a 15-stage gas turbine compressor, the platform resulted in a 0.37 percentage points efficiency improvement, reducing parameterized variables to 32. Utilizing isentropic efficiency as the pre-screening criterion with a population size of 200 and 560 true evaluations, the entire optimization process took 79.2 hours. This data-driven aerodynamic optimization platform demonstrates efficiency and stability in three-dimensional rapid parameterized optimization for highly loaded multi-stage compressors, presenting a novel approach for such optimization challenges.
Liu et al. (Mon,) studied this question.