Unsteady aerothermal asymmetry and control in rotor-stator cavities: A synergistic LES and CNN-LSTM framework

Abstract

Amidst global energy shortages and the pursuit of carbon neutrality, high-temperature rotating equipment such as gas turbines and aero-engines requires advanced thermal management technologies. This study focuses on the rotor–stator cavity (RSC) and employs large eddy simulation (LES) to systematically investigate the asymmetry and spatiotemporal evolution of heat flux under high-Reynolds-number conditions. The results reveal that the combined effects of centrifugal force, Coriolis force, and local vortical structures lead to pronounced non-uniformity in the thermal boundary layer along both the axial and circumferential directions, thereby inducing strong local temperature gradients and fluctuations in heat transfer coefficients. To overcome the limitations of conventional models in predicting complex turbulent heat transfer, a hybrid data-driven framework based on CNN-LSTM is developed. The model leverages convolutional neural networks to extract local spatial features and long short-term memory networks to capture temporal dependencies, achieving high-accuracy fitting of the time-series distribution of heat transfer coefficients on the rotor surface, with coefficients of determination exceeding 0.96. Its robust performance and adaptability under diverse turbulent conditions are thereby demonstrated. Furthermore, comparisons with experimental data confirm that LES can effectively reproduce the intricate heat flux structures within the RSC, while the CNN-LSTM model provides reliable support for rapid prediction and engineering optimization. Overall, this research deepens the understanding of heat transfer mechanisms in rotating flow fields and establishes a theoretical and technical foundation for intelligent cooling control and high efficiency thermal management, with promising potential for aerospace engineering applications.