Flow characteristics prediction and sealing performance optimization of labyrinth seals in organic Rankine cycle with deep learning

The low-grade energy utilization technology based on the organic Rankine cycle is an effective approach to solving global climate issues. Labyrinth seals in turbines can reduce fluid leakage and improve system efficiency. This paper presents a comprehensive evaluation of the performance of labyrinth seals with the thermo-fluid-structural coupling method. Employing deep learning methods, the flow characteristics prediction and sealing performance optimization of labyrinth seals are achieved. The results demonstrate that the arrangement of the disturbance groove on the stator wall can enhance energy dissipation in cavities and significantly reduce leakage, among which the circle groove performs the best. The well-trained deep learning model can reconstruct the physical distributions of cavities within 3 ms, and accurately recognize the leakage and thrust force, with prediction accuracy much higher than traditional surrogate models. The optimization framework based on a deep learning model and a gradient-based algorithm achieves great convergence and strong robustness in all optimization tasks. After optimization, the leakage of the circle groove seal is reduced by up to 19.63% compared to the straight seal structure. This research provides crucial references for further performance improvement of organic working fluid labyrinth seals, and also practical cases for the widespread application of deep learning.