The aerodynamic performance of multi-stage axial flow turbines is critical for advancing gas turbine efficiency and reducing carbon emissions. However, conventional optimization approaches rely heavily on computationally expensive CFD simulations, limiting their practicality for multistage systems. This study develops and validates a two-round optimization framework that integrates a rapid meanline solver, a multi-objective genetic algorithm, and multiple data mining techniques. In the first-round, all design variables are optimized using the meanline model, generating a comprehensive design space. Four complementary data mining methods—DACE-Kriging, parallel coordinate analysis, self-organizing maps, and total variation analysis—are then applied to identify and cross-validate the most influential variables. A second-round optimization is subsequently performed, restricted to these key variables. The methodology is demonstrated on a four-stage axial turbine, with performance improvements validated by CFD simulations. Results show a 1.34% efficiency gain from the first-round optimization and an additional 0.89% from the second, yielding a total improvement of 2.24% over the baseline design. Remarkably, the entire optimization process is completed within 2 days, compared with the months required by traditional CFD-based methods. These findings highlight the novelty and effectiveness of integrating data mining into meanline-based optimization, offering a robust, efficient, and generalizable tool for turbine design.
Yang et al. (Mon,) studied this question.