Rapid prediction of particle deposition characteristics on typical turbine blade surface based on deep learning

In response to the issue of particle deposition on turbine blades in dusty environments, which affects the performance and lifespan of aero-engines, this paper presents a rapid prediction method for the characteristics and influencing patterns of particle deposition on turbine blade surfaces using Convolutional Neural Networks (CNNs). First, numerical simulations of particle deposition on typical C3X-type turbine blades are conducted using the critical velocity model based on EI-Batsh to establish a dataset. Second, a rapid prediction model using the U-Net network architecture is constructed. The rapid prediction model is then validated. Finally, this model is used to investigate the influence of particle diameter on deposition patterns. The results indicate that the CNN model developed in this study exhibits high accuracy and generalization ability. Using pixel difference and structural similarity as indicators, the maximum error in the training results is 0.04, with a minimum of 0.03. The maximum error in the prediction results is 0.22, with a minimum of 0.06. This verifies that the rapid prediction model demonstrates high accuracy and generalization ability, and can quickly reconstruct CFD calculation results. Therefore, the trained CNN model can replace CFD simulations in subsequent optimization processes.