Session: 05-06 Topics in Instrumentation (II)

Paper Number: 81965

81965 - Experimental Investigation on the Accuracy of Blade Tip Timing Spectral Analysis Methods 

Blade tip timing (BTT) is a powerful approach to measure the deformation of rotating blades in a duct. Compared to the conventional strain gauges (SG), BTT enjoys some unique advantages such as being non-intrusive, easy to deploy, and capable of long-time online monitoring all the blades in a row. One of the featured applications of BTT is to measure the vibrational characteristics of individual blades. The main challenge for this task is to overcome the frequency aliasing rooted in the under-sampling nature of BTT systems. A number of algorithms have been developed to address this problem. However, most of these works focus on the ability to recover dominant frequencies and little attention has been paid to the accuracy of the recovered amplitudes. In fact, the amplitude of displacement is critical for engineers to judge whether the vibration of a certain mode is over the limit.

In this paper, the mode recognition performances of some popular algorithms including non-uniform discrete Fourier transformation (NUDFT), compressed sensing (CS), multiple signal classification (MUSIC), and minimum variance spectrum estimate (MVSE) are investigated. The accuracy of amplitude recognition is highlighted. To accomplish this, the key is to acquire the real displacement of the blade tip. We devised a novel experimental facility where laser displacement sensors (LDS) are fixed on the rotor to monitor the blade movement and the signal from LDS is transferred through a slip ring. The application of LDS allows us to directly validate the precision of raw displacement measurement from BTT which ensures good quality of the input for the mode recognition algorithms. Meanwhile, traditional periodograms based on LDS data could provide a reliable reference to examine the BTT methods.

It is discovered that the output of the candidate algorithms can be correctly interpreted only when suitable prior knowledge is provided. For example, the NUDFT exhibits severe aliasing due to the lack of sensors. Nevertheless, the amplitudes at dominant frequencies are 80% accurate, higher than the other methods. Therefore, it is hard to understand the NUDFT spectrum without a Campbell diagram. MVSE produces better results than NUDFT once the frequency bands are correctly estimated. Thus, reliable finite element analysis should be conducted before using MVSE. The CS method yields neater spectrum with less prior information while the amplitudes are approximately 50% lower than the real value. The significant difference indicates that a detailed calibration must be performed before this method is used independently. The performance of MUSIC is not as good as expected. Its stability is further reduced when the signal-to-noise (SNR) ratio is low.

Considering that sensor breakdown is not rare in field tests, its effect on the algorithms’ performance is also studied. It is found that the difficulty of recovering non-synchronous vibrations (NSV) is lower than synchronous vibrations (SV). For NSV, we could use samples from several continuous cycles to include more information in a batch. The accuracy can thus be elevated by 10-30%. For SV, such a technique is not helpful. In this case, NUDFT displays the best stability.

The results and conclusions in this work could provide a helpful reference for researchers to process data obtained in rig tests. The advantages as well as disadvantages of NUDFT, CS, MUSIC, and MVSE are discussed. The importance of prior knowledge in accurately interpreting BTT results is highlighted. More sophisticated algorithms with fewer dependencies are expected in the future.

Presenting Author: Zhicheng Xiao Shanghai Jiao Tong University

Presenting Author Biography: My name is XIAO Zhicheng. I graduated from Shanghai Jiao Tong University in 2018 with a Bachelor’s Degree. Now I am a PhD student. My research interest is computational fluid dynamics and aeroelasticity in turbomachinery.

Authors:

Zhicheng Xiao Shanghai Jiao Tong University
Yiming Meng Shanghai Jiao Tong University
Hua Ouyang Shanghai Jiao Tong University