| description abstract | Abstract. The energy conversion of aero engines is borne by turbines, which are closely related to engine performance and service life. In order to operate stably in harsh conditions and high-temperature inflow environments, sophisticated cooling structures are needed to help reduce blade temperature, alleviate thermal stress, and avoid erosion. However, there are diverse uncertainties throughout the design, manufacturing, and operational processes, which can have potential negative impacts on performance dispersion and life predictions. This study evaluates the impact of turbine parameters on the coupled fluid–solid thermal transfer analysis and proposes an optimization strategy based on UQ (uncertainty quantification) to improve turbine cooling performance. A typical fully cooled turbine was used in the study, with features such as inner coolant passage, impingement cooling, film cooling, rib walls, and pin fins. An automated design platform was created, allowing designers to easily modify parameters and assess aerothermal performance. This platform integrates 3D modeling, meshing, computational fluid dynamics (CFD) calculations, postprocessing, and UQ analysis. With the help of machine learning, a deep neural network was used to train a robust and efficient surrogate model. Based on SHAP (Shapley additive explainable) values, an adaptive dimensionality reduction method was developed by selecting the most sensitive geometric parameters, making the high-dimensionality analysis possible, as well as turbine optimization. The research results show that uncertainty assessment and optimization are of great significance for improving cooling performance and turbine reliability. | |
