Session: 40-02: Axial Compressor Instabilities and Stall
Paper Number: 152088
Surge and Rotating Stall Detection for Axial Compressor Flow Based on Self-Organizing Map
Aircraft engines must satisfy stringent safety standards. In the compressor, high-pressure ratio conditions can cause unstable flow phenomena such as rotating stall and surge. These unstable flows place a significant load on the blades and pose high safety risks. Currently, operational safety is ensured by maintaining a margin from the surge line to prevent dangerous conditions. However, not only must adequate surge margins be ensured, but also rotating stalls and surges must be predicted accurately to enhance aircraft safety. This study uses a self-organizing map (SOM), which is a type of unsupervised neural network, to predict surge and rotating stall in an axial compressor and estimate the appropriate surge margin.
A four-stage axial compressor rig test was conducted at Kawasaki Heavy Industries, Ltd., with corrected rotating speeds set to 85% and 90% of the designed condition. Unsteady pressure measurements were performed on the casing wall immediately upstream of the first-stage rotor blades at each rotational speed. During measurements, the mass-flow rate was gradually reduced by closing the exhaust valve until a surge event was observed. The onset of rotating stall and surge were observed at each corrected rotational speed.
The SOM was generated using the measurement data, which included rotating stalls and surges. The SOM, which employs unsupervised learning, visualizes similarities between high-dimensional data in a low-dimensional space. Pressure waveform data were partitioned into 20-rotation segments, transformed into the frequency domain using Fast Fourier Transform, and used to train the two-dimensional SOM. The results show that the SOM clearly classified regions of stable operation, rotating stall onset, and pre-surge conditions. Additionally, the data were mapped with a gradient-like transition corresponding to the time of data measurement such that the SOM included temporal information. Furthermore, when the test data not used in the training were input into the SOM, the best matching unit, i.e., the point on the map with the closest Euclidean distance to the input data, was identified on the map. The test data points were placed at positions that corresponded to the operating conditions of the compressor. This suggests that the SOM, which was generated from prior measurement data, allows one to estimate the internal flow state of the compressor in real time. Because the method can classify not only rotating stall but also the data immediately before a surge occurs, it enables the accurate prediction of surge occurrence, thus improving aircraft safety.
Presenting Author: Hironori Miyazawa Tohoku University
Presenting Author Biography: Hironori Miyazawa completed his Ph.D. in Information Sciences from Tohoku University in March, 2019. Since April of the same year, he has been an Assistant Professor in the Department of Computer and Mathematical Sciences at the Graduate School of Information Sciences, Tohoku University. His main research interests are multiphysics numerical simulations of unsteady flows through turbines and axial compressor blade rows.
Authors:
Hironori Miyazawa Tohoku UniversityYoshihiro Kitamura Tohoku University
Takashi Furusawa Tohoku University
Satoru Yamamoto Tohoku University
Kentaro Nakayama Kawasaki Heavy Industries, Ltd.
Naoyuki Niwa Kawasaki Heavy Industries, Ltd.
Naoki Kanazawa Kawasaki Heavy Industries, Ltd.
Surge and Rotating Stall Detection for Axial Compressor Flow Based on Self-Organizing Map
Paper Type
Technical Paper Publication