Vol. 44 Supp. SCIENCE IN CHINA (Series A) August 2001 Acoustic emission technique based rubbing identification for Rotor-bearing systems HE Yongyong, CHU Fulei & WANG Qingyu State Key Laboratory of Tribology, Department of Precision Instruments, Tsinghua University, Beijing 100084, China Received July 5, 2001 Abstract Rubbing is the frequent and dangerous fault in the rotating machine, and efficient identification of the rubbing is a hot research subject in the field of fault diagnosis. In this paper, a new rubbing identification method is proposed, which is based on the acoustic emission technique. In this method, the acoustic emission signal of the rubbing in the multi-support rotor-bearing system is acquired by the acoustic emission sensor, and then the continuous wavelet transform is utilized to analyze this signal. Based on the rubbing mechanism, the frequency feature of the multiple frequency relation in the instantaneous frequency wave is extracted as the rubbing identification feature. The experimental results prove that the proposed method is efficient and feasible. Keywords: rubbing, acoustic emission, wavelet transform. The requirement of high efficiency and compactness has made the operation clearance of the modern rotating machinery become smaller and smaller, which will increase the probability of the rubbing between the stationary and rotational parts and thus may cause serious malfunction [1]. Therefore, the fault diagnosis of the rubbing is very important to the safe operation of the rotating machinery. A comprehensive research has been performed on the vibration of a rub-impact rotor system. But it has been proved that the vibration based rubbing identification is not efficient and misdiagnosis is very severe. The reasons should be that the vibration signal always contains high level noise and is always contaminated by other faults [2]. Acoustic emission (AE) is the phenomena of the strain energy emission via elastic wave when the particles in a mass (such as atom, molecule, etc.) move relatively [3]. It has been proved that the rubbing between the stationary and rotational parts would cause acoustic emission because of the elastic strain at the rubbing location, and this acoustic emission signal contains abundant rubbing information. In addition, different from the vibration signal, the amplitude of AE signal only depends on the energy created by the rubbing and is independent of the vibration of the rotor-bearing system [4]. It means that the vibration noise and other faults can not contaminate the AE signal of the rubbing, and it should be very pure. From this point, an AE based rubbing identification method is proposed, which includes two aspects: AE signal acquisition and the rubbing characteristic extraction. The AE signal of the rubbing is acquired by the AE sensor which is mounted on the side of the bearing housings in the radial direction. The continuous wavelet transform (CWT) is utilized to analyze the AE signal and extract the frequency feature of the rubbing in the instantaneous frequency wave of CWT to identify the rubbing, which is based on the rub-
460 SCIENCE IN CHINA (Series A) Vol. 44 bing mechanism. The paper is organized as following: In the section 2, the experimental setup is described. Section 3 introduces the AE signal firstly, and then gives out the comparison analysis between the AE signal and the vibration signal of the rubbing. Section 4 presents a detailed description of the proposed rubbing identification method and the experimental results. The conclusion and discussion are given in the last section. 1 Experimental setup The experimental setup is a rotor-bearing system consisting of four supports and four discs, as shown in fig.1, and with a set of data acquisition system including AE sensors. The four bearings are hydrodynamic lubricating bearings with the same inner diameter of 31mm. The length of the bearing housing in the axial direction is 40mm. The four discs are approximately symmetric along the solid coupling, and the mass of each disc is 2.205kg. The diameter of the axle is 12 mm. Fig. 1. The experimental Rotor-bearing systemt. 1. Motor; 2. flexible coupling; 3, 1st bearing housing; 4, discs; 5, keyphasor; 6, 2nd Bearing housing; 7, solid coupling; 8, 3rd bearing housing; 9, rub screw; 10, 4th bearing housing. The natural frequencies of the rotor system are obtained by impulse test, which are 56, 61, 154 and 195 Hz respectively. The speed of the 120 W motor is set at 4000 rpm and the rotating frequency of the rotor is about 4000/60 = 66.7 Hz, just between the second and the third natural frequencies. The rubbing screw is made of copper and the axle is steel material. The AE signal acquisition system is BUAA. One AE sensor and one acceleration sensor are mounted on the two sides of the third bearing housing in the radial direction. 2 Signal acquisition 2.1 Acoustic emission signal In the late 1940s, it is Josef Kaiser who first proposed using acoustic emission to detect the defect growth. He found that engineering materials emitted low amplitude clicks of the sound when they were stressed. In addition to AE sources associated with defect growth (i.e. plastic deformation and crack extension), the AE sensors are also found to be sensitive to the plethora of
Supp. ACOUSTIC EMISSION TECHNIQUE BASED RUBBING IDENTIFICATION 461 other energy loss mechanisms such as impacts, friction, turbulence, cavitation, spalling, material reduction etc. Now, The AE technique has become a powerful method for nondestructive testing (NDT) [5]. But little work has been done on the AE based rubbing identification in the field of the fault diagnosis. Different from vibration signal, the AE signal is converted from the sound produced when the rubbing occurs. Therefore, it is more related to the condition of the rubbing other than the vibration behaviors of the rotor [6]. If no rubbing between the rotor and stator occurs, the sensors usually have no response to such influences as imbalance and misalignment, which are difficult to be removed and will induce higher harmonics in the analysis of the vibration signals of a multi- support rotor system. Therefore, the AE technique is especially suitable for the rubbing identification in the multi-support rotor system, such as steam turbines. 2.2 Signal acquisition and comparison analysis The sample frequencies of two sensors (as aforementioned AE sensor and acceleration sensor) are all 18182Hz, and the sample number is 2048. Fig. 2 gives out a set of the time domain wave of the rubbing acquired by the AE sensor and vibration sensor respectively. These two sorts of signal are based on different mechanism and their physical attributes are not identical. Therefore, they should be compared with each other not in the time domain but in the frequency domain. Fig.3 gives out their spectrum graphs by FFT (Fast Fourier Transform: FFT). Amp. Amp. Time (s) Time (s) Fig. 2. The time waves of the rubbing. The AE signal; the vibration signal. It can be seen from fig.3 that the computed rotating frequency is 71 Hz, and the multiple frequency relation of the rotating frequency, its double and its triple, can be observed clearly. It means that the spectrum graph of the AE signal of the rubbing can present explicit rubbing characteristic according to the rubbing mechanism. Whereas, the spectrum in fig. 3 is much disordered and contaminated by high level noise, i.e., it is somewhat difficult to obtain the rubbing characteristic from the spectrum of the vibration signal of the rubbing. This fact also demonstrates that using AE to detect the rubbing in the rotor-bearing system is reasonable.
462 SCIENCE IN CHINA (Series A) Vol. 44 3 AE signal based rubbing identification 3.1 Wavelet analysis of the rubbing signal [7] Generally, the signal acquired from a stochastic process can be described in time domain and frequency domain respectively. FFT is the classical measure for describing the frequency domain characteristic of the signal, and has proved very efficient and convenient in the engineering field. Whereas, FFT almost does not provide any time domain information of the signal, and thus is only suitable for the stationary signal. As we know, rubbing in the rotor-bearing system is a typical time-varying and non-stationary process, thus, continuous wavelet transform, a powerful and popular time-frequency method, should be used to analyze the rubbing signal and extract rubbing characteristic for the identification. Fig. 3. The spectrum graphs of the rubbing signals. The spectrum of AE signal; The spectrum of vibration signal. By means of CWT, we have carried the analysis upon the rubbing signals presented in fig.2, and fig.4 gives out the corresponding instantaneous frequency wave at some time point, and the amplitude of the wave is just the modulus of the wavelet transform. It can be seen from fig.4 that the computed rotating frequency is 68 Hz (the real rotating frequency is 66.7 Hz), and the multiple frequency relation of one time, two times and three times can be also observed clearly, but the results are more accurate than that in fig.3. The reasons are discussed as following: In our experiment, considering the sample accuracy and the store memory, the sample frequency is set so high as 18182 Hz and the sample number is relative small (2048). If the data is not supplemented by interpolation, the frequency resolution by FFT should be 18182/2048=8.88 Hz. Therefore, with such frequency resolution, the rotating frequency computed from FFT would not be satisfactory. On the other hand, although the frequency resolution by CWT at any scale of a would not exceed that by FFT, the frequency difference between two scales can be very small and varies with the scale. For example, for a 1 =6 and a 2 =5, f=6.12 Hz; for a 1 =218 and a 2 =217, f = 0.3 Hz. It means that we can obtain higher frequency resolution by CWT than by FFT. In addition, from fig.4 we can also see that the instantaneous frequency wave of the vibration signal still does not present the rubbing characteristic of the multiple frequency
Supp. ACOUSTIC EMISSION TECHNIQUE BASED RUBBING IDENTIFICATION 463 clearly, and it also proves that using AE signal to identify the rubbing is reasonable and feasible. Fig. 4. The instantaneous frequency wave of AE signal at t = 0.009955 s; the instantaneous frequency wave of vibration signal at t = 0.03586 s. t/s Fig. 5. The instantaneous frequency graph by maximum-modulus method. 3.2. AE signal based rubbing identification method The multiple frequency relation described in fig.4 does not always hold at any time point by CWT due to high level noise. If the noise is severe, the time point at which such multiple frequency relation exists will be much fewer. Therefore, it will be difficult to discover and distinguish such relation by general maximum-modulus method of CWT (depicted as fig.5). In addition, we cannot observe such relation point by point through all the time (the sample number is 2048 in our experiment and one data corresponds to one time point). So, a standard spectrum model is designed which is of multiple frequency relation, and used to match the computed instantaneous frequency wave by CWT point by point. In addition, a calculator is designed to calculate the number of the time point at which the computed instantaneous frequency wave is matched by the standard model, this number represents the rubbing characteristic and can be used to detect the rubbing. To verify the fact that the multiple frequency relation in the instantaneous frequency wave is induced by the rubbing mechanism, we design following experiment for comparison analysis: Except two rubbing signals presented in fig.2, another vibration signal without rubbing is acquired, and a stochastic signal is designed according to fig.3, the spectrum of which concen-
464 SCIENCE IN CHINA (Series A) Vol. 44 trates within 0 400 Hz. The spectrum graphs of these two signals are presented in fig.6. Then, we can carry out the comparison study with these four signals (the AE signal and vibration signal with rubbing, the vibration signal without rubbing and stochastic signal) by the proposed method, and table 1 gives out the results. From table 1, we can see that the signals with rubbing can be distinguished from the signals without rubbing and the stochastic signal by the statistic number of the multiple frequency relation in the instantaneous frequency wave. It also proves that such frequency feature is induced by the rubbing mechanism. In addition, it can be seen that the AE signal of the rubbing contains most time points and far more than other signal under all control error. Therefore, we can use this frequency feature by CWT of the rubbing AE signal to identify the rubbing, and the proposed method is efficient and feasible. Fig. 6. The vibration signal without rubbing; the stochastic signal. Table 1 The statistic number of the time point with multiple frequency relation (2048 points Signal Multiple frequency relation of 1, 2 Multiple frequency relation of 1, 2, 3 0.01 * 0.02 * 0.03 * 0.03 * 0.04 * 0.05 * AE (with rubbing) 70 138 496 35 62 103 Vibration (with rubbing) 27 67 190 14 29 63 Vibration (without rubbing) 2 14 21 0 0 1 Stochastic 14 30 69 2 3 15 * Control error for pattern matching. It needs to be regarded that the control error, which is used during the matching between the standard model and the computed wave, should be determined carefully. If the error is too big, misjudge will increase, especially for stochastic signal, because the spectrum of the stochastic signal distributes on all frequencies. If the error is too small, the rubbing characteristic of the multiple frequency relation cannot be extracted sufficiently and some information will be lost. In our experiment, six control errors and their corresponding results are given (presented in table 1). In the application, the control error should be determined according to the statistic analysis upon the historical running data of the equipment, and make the control error more suitable for the specific equipment.
Supp. ACOUSTIC EMISSION TECHNIQUE BASED RUBBING IDENTIFICATION 465 4 Conclusions Rubbing between the rotor and stator in the rotor-bearing system can create acoustic emission. Using this attribute of the rubbing, the AE based rubbing identification method is proposed, which detects the multiple frequency relation in the instantaneous frequency wave of CWT through pattern matching, and use this frequency feature of the rubbing AE signal to identify the rubbing. The experimental results show that the proposed method is efficient and feasible. Acknowledgements ThisworkwassupportedbytheOpeningFoundationofStateKeyLab.ofVSNofShanghai Jiaotong University (Grant No. VSN-2001-02). References 1. Huang Wenhu, Xiao Shongbo, Liu Ruiyan, Equipment Fault Diagnosis Theory, Technique and Application, Beijing: Science Press, 1997. 2. Zhang Aiping, Sun Wei, Ye Rongxue. Study on bearing monitoring for turbine unit based on acoustic emission, TurbineTechnology, 1998, 40(1): 29 32. 3. Holroyd, T. J., Acoustic emission an NDT technique evolving into a versatile industrial monitoring method, Measurement & Control, 1997, 30: 141 145. 4. Wu, J., Liang, J., Li, H., Acoustic emission apparatus for rubbing diagnosis of large rotating machinery, Journal of Beijing University of Aeronautics and Astronautics, 1998, 24: 104 107. 5. Liang, S. Y., Dornfeld, D. A., Tool wear detection using time series analysis of acoustic emission, Transactions of the ASME, Journal of Engineering for Industry, 1989, 111: 199 204. 6. Belyi, V. A.,, Kholodilov, O. V., Sviridyonok, A. I., Acoustic spectrometry as used for the evaluation of tribological systems, Wear, 1981, 69: 309 318. 7. Mallat, S., A theory for multiresolution signal decomposition: The wavelet representation, IEEE Transactions On Pattern Analysis And Machine Intelligence, 1989, 11: 674 693.