A Corner Separation Sensing Method Based on Shallow Neural Networks in Compressor Cascade

In many applications, it is important to obtain real-time flow state from limited sensors for the development of closed-loop flow control technology. In this work, we propose a method that combines a shallow neural network with a corner separation factor for rapidly detecting corner separation in a compressor cascade to address the requirements of closed-loop control. Our approach uses a shallow neural network to quickly reconstruct the pressure distribution on the suction surface of the cascade using nine sensors, from which a corner separation factor, S_int, is derived to quantify the corner separation. The dataset comes from a large set of numerical simulations conducted under various operating conditions, including turbulence intensity, Mach number, angle of attack, and boundary layer profiles. The results show that the S_int accurately represents variations in corner separation losses across these conditions, with a Pearson correlation coefficient of 0.9423, indicating a strong linear correlation. A leverage score method is employed for sensor placement, where sensor locations are randomly selected based on score probabilities. Not only are the characteristics of separated flow considered, but it can also enhance the model's global perception capability. We also discussed the model's sensitivity and general performance under varying angles of attack. In addition, experimental validation was performed under varying angles of attack and Mach numbers. The results show that this method can capture changes in flow field and effectively assess the corner separation. This research demonstrates the potential of the proposed method for rapidly assessing corner separation in compressor cascades, supporting the advancement of closed-loop flow control technology.