Deep learning-driven study on thrust deflection suppression in serpentine nozzles via blowing flow control

The serpentine convergent–divergent nozzle combines stealth and high-performance flight, making it a promising exhaust concept for sixth-generation fighters. However, under overexpanded conditions, asymmetric flow separation can induce thrust deflection and threaten flight stability. This study investigates the physical mechanisms and parametric effects of blowing flow control using an integrated CFD–deep learning approach. A regression–classification joint model based on the Multi-Layer Perceptron (MLP) is constructed to classify the effectiveness of thrust deflection suppression under different combinations of Nozzle Pressure Ratio (NPR), Blowing-to-Back Pressure Ratio (BBPR), and blowing position. The prediction results reveal a three-stage evolution of the successful suppression region with increasing NPR: a predominantly isolated success stage (NPR≈1.4–1.6), where strongly asymmetric separation makes effective suppression rely on high-BBPR momentum input, often at the cost of reduced flow capacity; a dual-region control stage (NPR≈1.6–1.75), where the dual-λ shock and the curved shock between the throat and slot correspond to lower-left (low BBPR with upstream blowing) and upper-right (high BBPR with downstream blowing) success regions; and a fully connected stage (NPR≈1.75–2.1), where the dual-λ shock becomes dominant and suppression is maintained over a broad parameter range except for extreme blowing configurations.