Acoustic emission based double impulses characteristic extraction of hybrid ceramic ball bearing with spalling on outer race

Transcription

1 Acoustic emission based double impulses characteristic extraction of hybrid ceramic ball bearing with spalling on outer race Yu Guo 1, Tangfeng Yang 1,2, Shoubao Sun 1, Xing Wu 1, Jing Na 1 1 Faculty of Mechanical and Electrical Engineering Key Lab. of Vib. & Noise under Ministry of Education of Yunnan Province Kunming University of Science and Technology, 655, Kunming, Yunnan, China kmgary@163.com 2 Technology Center of Kunming Yunnei Power Co., Ltd Kunming Yunnei Power Co., Ltd, 65224, Kunming, Yunnan, China Abstract Double impulses phenomenon can be observed in vibration which is generated by every rolling element passing over a spall zone on the outer or inner races of a rolling element bearing. Acoustic emission (AE) technique has been incorporated into the incipient defect detection of bearing heath for many years. However, whether or not the double impulses phenomenon of a spalled hybrid ceramic ball bearing can also be observed in AE waveform and its characteristics have never been reported. In this paper, the double impulses phenomenon of faulty hybrid ceramic ball bearings with spalled outer races has been investigated. The study shown that the double impulses phenomenon can also be observed in the AE waveform. The separation treatment is used to extract the double impulses from the AE signal in this paper. Experiment results indicate that the double impulse phenomenon can be effectively extracted from the AE signals generated by hybrid ceramic ball bearings with spalls on outer races. 1 Introduction Rolling element bearing (REB) is widely used in rotational machinery. Spalling is the main cause of the failure of REBs. It is worth mentioning that even the incipient spalling occurs, the REB can still work for a long time and a premature replacement of the bearing can be costly [1]. Today, hybrid ceramic ball bearings (the rolling ball made by ceramic materials, such as Si 3 N 4 ) are widely utilized in high speed rotating machinery. However, the potential failure is one of the biggest causes of machinery breakdowns. Then, the estimation of the remaining useful life (RUL) of such hybrid ceramic ball bearings is important. It is known that double impulses phenomenon can be observed in vibration which is generated by every rolling element passing over a spall zone on the outer or inner races of a rolling element bearing. Some literature reports that double impulses phenomenon can be employed to estimate the spall length along the rotating direction of the faulty ball bearing by measuring the space between the double impulses [1]. This phenomenon can be helpful for the evaluation of the RUL of hybrid ceramic ball bearings. Acoustic emission (AE) technique has been incorporated into the incipient defect detection of bearing heath for many years [2]. However, whether or not the double impulses phenomenon of a spalled hybrid ceramic ball bearing can also be observed in AE waveform has not been reported in the literature. In this paper, the double impulses phenomenon of faulty hybrid ceramic ball bearings with spalled outer races is investigated. 2447

2 2448 PROCEEDINGS OF ISMA216 INCLUDING USD216 The study shown that the double impulses phenomenon can also be observed in the AE waveform when every ceramic ball passes over the spall-type area of a faulty bearing. Since the AE signal is more sensitive to the incipient fault and the sampling ratio generally can reach to megahertz, a clearer double impulses phenomenon can be observed in a high frequency band by isolating the potential noise. The paper is arranged as follows. Firstly, the double impulse phenomenon of an outer race spalled REB will be introduced in Section 2. Subsequently, the AE based double impulse phenomenon extraction approach will be presented in Section 3. Then, experiments will be shown in Section 4. Finally, conclusions will be drawn in Section 5. 2 Double impulses phenomenon of an outer race spalled REB 2.1 Explanation on double impulses phenomenon The schematic of the rolling element passing over a spall-like area can be shown in Fig. 1. F N represents the force reaction of the raceway to the rolling element, and V is the tangential velocity of the rolling element. Points A and B indicate the entering point and exiting point in the process of rolling element passing over FN FN FN V V V A B C (b) (c) Figure 1: Schematic of a ball rolling over a spall region. the spalling zone, respectively. The entering event can be simply illustrated as the rolling element rolling from A to B, and the exiting event is the process of rolling element rolling from B to C. According to Sawalhi s previous work [1], the rolling element move over the leading edge of the spall region at point A. It is a de-stress progress, which generates a step response. On the contrary, when the rolling element rolling over point B, it is a re-stress progress. Meanwhile, an impact between the rolling element and the trailing edge of the spall region occurs, which generates an impulse response. Besides, previous work also indicates that the step response mainly contains low-frequency components and the impulse response is a wide-band signal. Because the two responses are with different characteristics, the separation scheme original proposed in [1] should be is more suitable to realize the double impulses extraction. It is worth pointing out that all the previous studies on the double impulses phenomenon related REB diagnosis techniques are based on vibration analysis. 2.2 Double impulses phenomenon in acoustic emission Even though the AE based REB diagnosis techniques have been developed in recent decades. But whether or not the double impulses phenomenon in spalled REB can be observed in the AE signal has not been reported yet. This study is inspired by the motivation to address this issue. In our study, a comparison of experiment results on an outer race spalled REB through vibration and AE signal has been employed to investigate the double impulses phenomenon in the AE signals. In this scheme, the test is carried on the QPZZ-II test rig, which is shown in Fig. 2. This test rig is driven by a three phase AC motor whose speed is controlled by a

3 MONITORING AND DIAGNOSTICS OF ROTATING MACHINERY 2449 Motor Loading device Bearing house Figure 2: QPZZ-II test rig. frequency converter. The power can be transmitted to the rig via a synchronous belt with the transfer ratio 1:1. Near the bearing house, there is a loading device to give the tested bearing a certain load downward. A deep groove ball bearing (type ZYS 625) with a 2 mm artificial spall width on the outer race, which is carved by the electro discharge machining (EDM), is tested. The picture of the spalled ball bearing is shown in Fig. 3. An eddy probe, as shown in Fig. 3(b), is mounted upon a rotating shaft and generates the tacho Eddy probe Bearing house Accelerometer 2mm (b) Acoustic emission transducer (c) Figure 3: Pictures of outer race spall, (b) eddy probe, and (c) accelerometer and AE transducers. pulses once per shat revolution by the keyway on the shaft. The tacho pulses can be utilized to calculate the speed profile of the reference shaft. The accelerator and the AE transducer, as shown in Fig. 3(c), are mounted at the almost same position on the bearing house, by which the potential influences on transfer paths related to the two kinds of signals can be reduced. An ADLINK PCI-9846H data acquisition card is use for data acquisition, and its sampling frequency is set to 1 MHz in the test to sample both the vibration and the AE signal synchronously. It is worth mentioning that the sampling rate 1 MHz used for picking the vibration may be too high in general applications of REBs diagnosis. However, this high sampling rate provides a convenience for the comparison with the AE signal in this study. The picked vibration and AE signal at 6 revolutions per minute (rpm) are shown in Fig.4 and (b) respectively. In the vibration plot, the step response and impulse response can be observed clearly. It is noted that some obvious oscillations are shown at the same positions corresponding to the step response and the impulse response respectively in the AE signal plot. Thus, it can be originally drawn a conclusion that the outer race spall related double impulses phenomenon can be observed in the AE signals. On the other hand, it is also noted that the waveforms of the entering event and the exiting event in AE signals are different from that in vibrations, where the double impulses corresponding oscillations are noiseless and smoothness. In this investigation, the two responses in the AE will be named as the entry response and the exit response respectively. Generally, these characteristics

4 245 PROCEEDINGS OF ISMA216 INCLUDING USD216 of two responses in the AE signal make the space measurement between the two events convenient. (b) Acceleration (m s 2 ) Volts (V) 2 2 step response impulse response entry response exit response Samples Figure 4: Observed waveforms of vibration and (b) acoustic emission (spall width 2 mm) 3 Double impulses extraction 3.1 Separation analysis technique For an accurate space measurement, the double impulses should be enhanced to promote the SNR. There are two approaches presented in [1] viz. the joint treatment and the separation scheme. The joint treatment can be used to enhance the two events together. In contrast, the separation enhance the two events separately but more steps are needed. As aforementioned, the main frequency components included in step response and impulse response are different in the vibration. It should be the same to the entry response and exit response in the AE signal. Thus, the separation scheme should be a better choice to deal with the AE based double impulses phenomenon in this study. In the separation scheme, the peak positions of the entry response and exit response should be captured firstly. Then the sample at the position of 2% samples from the second peak between the adjacent double peaks are taken as separation point, and in each double impulses would find a separation point. Subsequently, starting at one separation point and setting 5% 8% samples between the adjacent separation points to zero will get the entry response. At last, taking the source double impulses acoustic emission minus the entry response will get the exit response. By using the separation scheme to separate the double impulses in the observed AE signal shown in Fig. 4(b), the results are shown in Fig Feature enhancement Due to the oscillation of the waveforms, the two events related peaks in AE waveforms are not suitable for space measurement. Refer to [1], measuring the space of the two responses by the envelope waveform should be a better choice. However, to improve measurement accuracy, the SNR of the two events in the AE signal should be enhanced by some pre-processing steps. In this study, the Autoregressive (AR) model based pre-whitening method is used to eliminate the deterministic components at first. Assuming x t is the measured AE signal, the AR model based pre-whitening can be expressed as in [3, 4] by p e k = x k a j x k j, (1) j=1

5 MONITORING AND DIAGNOSTICS OF ROTATING MACHINERY (b).5.5 (c) Samples Volts (V) Figure 5: Waveforms of observed acoustic emission, (b) entry response, and (c) exit response. where e k is the residual signal, p represents the model order, and a j, j = 1, 2, 3,..., p are the weighting coefficients. After the pre-whitening, the random components related to the spall become prominent in the signal. To further enhance the impulsive components contained in the residual signal, which are mostly related to the spall, the minimum entropy deconvolution (MED) [5] is adopted. The MED is a system identification method that can deconvolve the impulsive excitations from mixture signals. In principle, it is to find an inverse filter for identifying impulsiveness components in a complicated signal. The most important parameters of the MED filter is the iteration times and the filter length. In this study, the two parameters are determined by the maximum kurtosis value of the MED output y k. By combining the AR and the MED, the bearing fault related bursts can be enhanced effectively [5]. The schematic for feature enhancement of the separated AE signal is shown in Fig 6. k p j a x j k j k k Figure 6: Schematic of AR and MED based feature enhancement. 3.3 Optimal envelope extraction by wavelet kurtogram After the feature enhancement, the outer race spall related components become prominent. Then, the two envelopes of the two events are extracted respectively for an accurate space measurement of the two events in

6 2452 PROCEEDINGS OF ISMA216 INCLUDING USD216 the separation scheme. The envelopes are extracted by the wavelet kurtogram [4], which is briefly introduced as follows. As well known, the complex Morlet wavelet is defined as in [4, 6, 7] by ψ(t) = σ π e σ2 t 2 e j2πf t, (2) where σ is the waveform parameter, and f denotes the modulation frequency. Its Fourier transform can be given by Ψ(f) = Ψ (f) = e (π2 /σ 2 )(f f ) 2, (3) where Ψ(f) is the Fourier transform of ψ(t), the asterisk denotes the complex conjugate, and Ψ(f) = Ψ (f) since Ψ(f) is real. It is worth mentioning that the complex Morlet wavelet can also be explained as a digital filter with the centre frequency f and the bandwidth σ. In the wavelet kurtogram [4], the complex Morlet wavelet is employed to construct a set of tree-like octave filter banks with shifted centre frequencies and constant proportional bandwidths. And the Fourier transform of the mth filter at the ith filter bank constructed by the complex shifted Morlet wavelet can be expressed by Ψ i m(f) = e (π/σi m )2 (f f i m )2, m = (1, 2, 3,..., M), (4) where M = k(i)n is the total number of the filters (k(i) denotes the number of filters per octave at the ith filter bank, N is the number of octaves to cover the whole analysis band), f i m and σ i m represent the shifted centre frequency and the constant proportional bandwidth of the mth filter at the ith filter bank. The bandpass filtering can be calculated by the inverse Fourier transform, which is given as in [7] by w(σ i m, f i m, τ) = F 1 [X(f)Ψ i m(f)], (5) where X(f) is the Fourier transform of the analyzed signal x(t), w(σ i m, f i m, τ) denotes the bandpass filtered signal at center frequency f i m with bandwidth σ i m, and it actually represents the complex envelope since ψ(t) is an analytic filter (see Eq.(2)), operation F 1 represents the inverse Fourier transform. Then the spectral kurtosis (SK) of the data can be calculated by ( ) SK σm, i fm, i c i m = w(σm, i fm, i τ) 4 ( ) w(σm, i fm, i τ) 2 2, f i 2 m (6) where symbol denotes mathematical expectation operation viz. mean value calculation, c i m represents the squared envelope of the mth filter at the ith filter bank, c i m = w(σ i m, f i m, τ) 2. The optimal bandwidth σ o, center frequency f o, and the corresponding squared envelopes c o can be determined by Eq.(6) with the maximum SK value. The procedures can be expressed as in [8] by (σ o, f o, c o ) = argmax(sk(σ i m, f i m, c i m)), (7) where the notion argmax is denoted to obtain the parameters pertaining to the maximum SK value calculated by the Eq. (6), f o and σ o represent the optimal centre frequency and bandwidth for the bandpass filtering, respectively, c o denotes the extracted optimal squared envelope. See [4] for the details about the wavelet kurtogram scheme. 3.4 Schematic for the double impulses extraction Based on the aforementioned methods, a schematic for the double impulses extraction in this study is proposed and illustrated in Fig.7. And the main steps are listed as follows. Firstly, the picked acoustic emission is separated into the entry response and the exit response by the separation scheme.

7 MONITORING AND DIAGNOSTICS OF ROTATING MACHINERY 2453 Acoustic emission of outer race spalled REB Entry response and exit response separation Entry response Exit response AR Pre-whitening and MED filter Optimal envelope extraction by complex Morlet wavelet AR Pre-whitening and MED filter Optimal envelope extraction by complex Morlet wavelet Normalized & added together Extracted double impulses Figure 7: Schematic of double impulses extraction. Subsequently, the AR model based pre-whitening and the MED filtering are used to enhance the impulsive components in the entry response and the exit response. Then, the wavelet kurtogram is employed to extract the optimal envelope of the two parts respectively. After that, the squared envelopes of the entry response and exit response are low-pass filtered respectively due to the envelope mainly contains low frequency components. Lastly, the extracted two envelopes are normalized separately and added back together to obtain the double impulses. 4 Experimental analysis To verify the validation of the AE based double impulses extraction, experiments have been investigated on the test rig shown in Fig.3. The test conditions are presented in subsection 2.2. The space measurement of the faulty hybrid ball bearing with 2 mm spall width on the outer race at rotating speed 6 rpm is analyzed in the follows. After the separation, the inspected AE signal (see Fig.4(b)) has been separated into the entry response and the exit response shown in Fig.5. Then, the AR pre-whitening and MED filtering are utilized to enhance the fault feature components embedded in the entry response and exit response respectively. The enhanced entry response and exit response are shown in Fig.8. Subsequently, the optimal envelopes are extracted by the wavelet kurtogram scheme.kurtograms of entry response and exit response are shown in Fig. 9 and Fig. 1, where the optimal filtering bands (f o, σ o ) are ( Hz, Hz) and ( Hz, Hz), respectively. It verified that, the main frequency band of the entry response and the exit response are also different, and the separation scheme is needed. At last, the two optimal envelopes are low-pass filtered and normalized respectively and then added them together to obtain the double impulses, which are shown in Fig.11.

8 2454 PROCEEDINGS OF ISMA216 INCLUDING USD216 Volts (V).5.5 (b) Volts (V) Samples Figure 8: Enhanced entry response, and (b) exit response. SK max level7, f o = Hz, σ o =158.4Hz SK max level7, f o = Hz, σ o = Hz Level 4 Level , 6, 7, 8, 9, 1, Frequency (Hz) 7 5, 6, 7, 8, 9, 1, Frequency (Hz) Figure 9: Kurtogram of entry response. Figure 1: Kurtogram of exit response. 2 1 (b) (c) Amplitude step response impulse response Samples Figure 11: Extracted entry response, (b) exit response, and (c) double impulses. 5 Conclusions The investigation results proved that the double impulses phenomenon can be observed in the AE signal of outer race spalled hybrid ceramic ball bearings. Different from the double impulses phenomenon observed in

9 MONITORING AND DIAGNOSTICS OF ROTATING MACHINERY 2455 vibration, the entry response and exit response in the AE signal are noiseless and smoothness. The AE based double impulses characteristics of hybrid ceramic ball bearing can be extracted by the separation scheme for the space measurement. Acknowledgements The Project sponsored by National Natural Science Foundation of China (Grant No ). References [1] N. Sawalhi and R. Randall, Vibration response of spalled rolling element bearings: Observations, simulations and signal processing techniques to track the spall size, Journal of Mechanical Systems and Signal Processing, Vol. 25, No. 3, Academic Press (211), pp [2] J. R. Matthews, Acoustic emission, Gordon and Breach Science Publishers Inc., New York, Vol. 2: CRC Press, (1983), ISSN [3] C. Junsheng, Y. Dejie, and Y. Yu, A fault diagnosis approach for roller bearings based on EMD method and AR model, Mechanical Systems and Signal Processing, Vol. 2, No.2, Academic Press (26), pp [4] N. Sawalhi, Diagnostics, prognostics and fault simulation for rolling element bearings, PhD Dissertation, The University of New South Wales Australia, Australia (27). [5] N. Sawalhi, R. Randall, H. Endo, The enhancement of fault detection and diagnosis in rolling element bearings using minimum entropy deconvolution combined with spectral kurtosis, Mechanical Systems and Signal Processing, Vol. 21, No.6, Academic Press (27), pp [6] N. Sawalhi and R. B. Randall, Spectral kurtosis optimization for rolling element bearings, in IEEE, Proceedings of the Eighth International Symposium on Signal Processing and Its Applications, Sydney, Australia, August 28-31, Sydney (25), pp [7] N. Nikolaou and I. Antoniadis, Demodulation of vibration signals generated by defects in rolling element bearings using complex shifted Morlet wavelets, Mechanical systems and signal processing, Vol. 16, No. 4, Academic Press (22), pp [8] Y. Guo, J. Na, B. Li, and R.-F. Fung, Envelope extraction based dimension reduction for independent component analysis in fault diagnosis of rolling element bearing, Journal of Sound and Vibration, Vol. 333, No. 13, Academic Press (214), pp

10 2456 PROCEEDINGS OF ISMA216 INCLUDING USD216