Session: 36-03 Fan and Engine Noise

Paper Number: 103096

103096 - Application of Compressed Sensing Method With L1/2-Norm in Fan/compressor Mode Detection 

Noise is an important indicator of airworthiness certification of commercial aircraft, and fan/compressor noise of aeroengines is a major noise source. Fan/compressor noise can be decomposed into a series of azimuthal and radial modes in the engine duct, and then radiating to free space from the exit plane of the duct. Accurate detection of acoustic modes is of great significance for guiding the design of low-noise aeroengines. However, due to the generally large number of rotor and stator blades, the order of azimuthal modes could be generated so high that hundreds of sensors would be needed for the detection when using the conventional spatial Fourier transform. The existing compressed sensing method based on l₁-norm minimization can effectively reduce the number of sensors in measurement, but the success probability and noise immunity remain to be improved. In this paper, l_1/2-norm method is used to replace the traditional l₁-norm in compressed sensing, and the reconstruction performance and applicability of the iterative reweighted l_1/2-norm minimization algorithm are studied through numerical simulations and experiments. Numerical simulations show that the l_1/2-norm algorithm can effectually eliminate the spurious peaks in the l₁-norm mode spectrum. In addition, the dynamic range of the l_1/2-norm algorithm is higher than the l₁-norm, and the background noise level of the mode spectrum decreases as the number of iterations increases. Under different number of sensors, different sparsity, and different noise levels, the l_1/2-norm algorithm always attains a higher success probability than the l₁-norm. In robustness analysis, the reconstruction error of the l_1/2-norm algorithm is lower than that of the l₁-norm algorithm in the case of random absence of sensor signals, which shows stronger reliability. In the acoustic mode detection experiment on a cooling fan, 24 sensors were used to decompose the mode at the blade passing frequency (BPF). The sparsity of the reconstructed mode spectrum of l_1/2-norm algorithm was less than that of the l₁-norm algorithm, and the amplitude of the main mode order -1 was increased by 2.7 dB. After randomly removing two groups of sensor data, the deviation between results and original data is also found smaller by the l_1/2-norm algorithm. In the aerodynamic mode detection on an axial compressor test with two sensors damaged, the l_1/2-norm algorithm improved the main mode amplitude at the BPF and the rotating instability frequency (RIF), further validating the ability to enhance sparsity. Numerical simulations and experiments show that using l_1/2-norm instead of l₁-norm can significantly improve the success probability, noise immunity, and robustness of the compressed sensing approach, which indicates a stronger application value in fan/compressor mode detection.

Key words: compressed sensing; mode detection; l_1/2-norm; l₁-norm

Presenting Author: Zhaoyin Li Shanghai Jiao Tong University

Presenting Author Biography: Zhaoyin LI the B.E. degree in Machanical Engineering from Shanghai Jiao Tong University, Shanghai, China,in 2022. He is currently working toward the M.E. degree in Machanical Engineering in Shanghai Jiao Tong University, Shanghai, China. His research interests include aero engines, axial flow fan and duct detection.

Authors:

Zhaoyin Li Shanghai Jiao Tong University
Zeyuan Yang Shanghai Jiao Tong University
Pengfei Chai Shanghai Jiao Tong University
Zonghan Sun Shanghai Jiao Tong University
Jie Tian Shanghai Jiao Tong University
Hua Ouyang Shanghai Jiao Tong University