Machine-learning-based optimization of labyrinth seal geometry considering both leakage and windage heating performance

초록

Increasing attention has been given to improving both leakage control and thermal management in labyrinth seals, motivating the development of integrated design approaches for enhanced aero-thermal performance. This study establishes a machine-learning-based optimization framework for a stepped honeycomb labyrinth seal to simultaneously reduce leakage and windage heating. The framework combines computational fluid dynamics, surrogate modeling, and a genetic algorithm to investigate the influence of key geometric parameters on seal performance. The discharge coefficient (Cd) and windage heating coefficient (6) were adopted to quantify aerodynamic and thermal performance. A sensitivity analysis identified the dominant factors influencing Cd and 6, clarifying the trade-off between aerodynamic efficiency and viscous dissipation. Among the tested surrogate models, the Random Forest approach provided the most accurate and stable predictions, reproducing CFD results for the optimized configurations with deviations below 3 percent. The multi-objective optimization achieved a 9.9 percent reduction in Cd with only a 3.4 percent increase in sigma, confirming that balanced geometric modifications can mitigate the aerodynamic-thermal trade-off. The proposed framework offers a physically inter-pretable, data-driven design strategy for improving seal efficiency and reliability in high-speed and high-temperature turbomachinery.

키워드