Aerodynamic optimization of multi-stage axial turbine based on pre-screening strategy and directly manipulated free-form deformation

To address the challenges of multiple design variables, long evaluation times, and poor global search capability of traditional surrogate-assisted algorithms that rely on the complete substitution of accurate function evaluations in the turbine aerodynamic optimization process, a prescreening surrogate-assisted elitist preservation genetic algorithm (pre-SAEGA) optimizer is proposed. Pre-screening strategy in the pre-SAEGA can screen samples instead of directly estimating them, thereby reducing expensive evaluations in each generation. The directly manipulated freeform deformation (DFFD) method is applied to parameterized multi-stage axial turbines, and multi-degree-of-freedom flexible control is realized. Combining the pre-SAEGA with the DFFD method, a data-driven multi-stage axial turbine optimization platform is established. A two-stage axial turbine is the research object, and 44 design variables are selected for the combined optimization design of flow path and blade rows. The results show that the isentropic efficiency and flow rate improve by 1.33 % and 1.81 % respectively, and the pressure ratio decreases by 0.47 % at the turbine design point. The presented optimization platform not only improves the aerodynamic optimization effect but also significantly reduces the number of design variables and real evaluation samples, making it suitable for solving multi-stage turbine optimization problems with multiple degrees of freedom.