Surge and Rotating Stall Detection for Axial Compressor Flow Based on Self-Organizing Map

Abstract Aircraft engines are required to meet stringent safety standards, while unstable flow phenomena, such as rotating stalls and surge in axial compressors, pose substantial safety risks. Ensuring operational safety requires maintaining adequate surge margins and accurately predicting these phenomena. This study employs a self-organizing map (SOM), an unsupervised neural network, to predict rotating stalls and surges in an axial compressor and estimate appropriate surge margins. A four-stage axial compressor rig test was conducted with corrected rotational speeds of 85 % and 90 % of the design condition. Unsteady pressure measurements were performed upstream of the first-stage rotor blades as the mass-flow rate was gradually reduced until a surge occurred. The onset of rotating stall and surge was observed at each speed. Measurement data, including pressure waveforms partitioned into 20-rotation segments and transformed via Fast Fourier Transform, were used to train a two-dimensional SOM. The SOM visualized similarities in high-dimensional data, classifying regions of stable operation, rotating stall onset, and pre-surge conditions. Temporal information was also captured, with data mapped in gradient-like transitions reflecting measurement time. When tested with unseen data, the SOM identified corresponding operating conditions in real-time based on the closest Euclidean distance, enabling accurate internal flow state estimation. The method effectively classified pre-surge conditions, offering enhanced predictive capabilities for surge occurrence. These results demonstrate the potential of the SOM to improve compressor safety by accurately predicting unstable flow phenomena, thereby contributing to aircraft safety.