You are on page 1of 9

Time Synchronous Average Based Acoustic Emission Signal Analysis on Gear Fault Detection

Yongzhi Qu, Junda Zhu, David He


Department of Mechanical and Industrial Engineering University of Illinois at Chicago Chicago, IL USA davidhe@uic.edu

Bin Qiu
Guangxi College of Water Resources and Electric Power 99 Changgang Road, Nanning, Guangxi, China

Eric Bechhoefer
NRG Systems Hinesburg, VT USA erb@nrgsystems.com

Abstract Acoustic emission (AE) has been studied as a potential information source for machine fault diagnosis for a long time. However, AE sensors have not yet been applied widely in real applications. Firstly, in comparison with other sensors such as vibration, AE sensors require much higher sampling rate. The characteristic frequency of AE signals generally falls into the range of 100 kHz to several MHz, which requires a sampling system with at least 5MHz sampling rate. Secondly, the storage and computational burden for large volume of AE data is tremendous. Thirdly, AE signal generally contains certain nonstationary behaviors which make traditional frequency analysis ineffective. In this paper, a frequency reduction technique and a modified time synchronous average (TSA) based signal processing method are proposed to identify gear fault using AE signals. Heterodyne technique commonly used in communication is employed to preprocess the AE signals before sampling. By heterodyning, the AE signal frequency is down shifted from several hundred kHz to below 50 kHz. Then a low sampling rate comparable to that of vibration sensors could be applied to sample the AE signals. After that, a modified tachometer less TSA method is adopted to further analyze the AE signal feature. Instead of performing TSA on the raw signals, the time synchronous averaging of the first order harmonic signal is obtained and analyzed. With the presented method, no tachometer or real time phase reference signal is required. The TSA reference signal is directly obtained from AE signals. By examining the smoothness of obtained wave form, a noticeable discontinuity or irregularity could be easily observed for gear fault diagnosis. AE data collected from seeded fault tests on a gearbox are used to validate the proposed method. The analysis results of the tests have shown that the proposed method could reliably and accurately detect the tooth fault. Keywords - time synchronous average; tach-less; heterodyne; acoustic emission;

caught in an early stage, significant maintenance cost could be saved. Hence, acoustic emission technique began to attract researchers attention to machine health monitoring and fault diagnosis. Acoustic Emission (AE) is commonly defined as transient elastic waves within a material, caused by the release of localized stress energy. It is produced by the sudden internal stress redistribution of material because of the changes in the internal structure of the material. Possible causes of these changes are crack initiation and growth, crack opening and closure, or pitting in various monolithic materials (gear, bearing material) or composite materials (concrete, fiberglass). Thus the ability to detect AE can be used to give diagnostics indications of component health. In comparison with vibration analysis, AE has the following advantages [2]: 1) insensitive to structural resonance and unaffected by typical mechanical background noises; 2) more sensitive to activities of faults; 3) provide good trending parameters; and 4) localization of measurements to the machine being monitored, i.e., AE signals are sensitive to the location of the faults. Challenge of using AE sensor include: The output signal from AE sensor are generally high frequency, even as high as several MHz. Thus very high sampling frequency is required, 5M-10MHz. Other challenges include the high data volume, complicated feature of AE signal, which make the data processing highly difficult. Until now, only limited techniques have been successfully applied to AE signal analysis. Traditionally, only the time domain features, like peak, total energy, standard deviation, median, AE counts, root mean square (RMS) voltage and duration have been extracted in condition monitoring [3][4]. However, these are all related to the absolute energy levels of the measured signal. As the absolute energy could vary from one machine to another, or vary in different locations on the same machine, so these kinds of standard could be inaccurate. Consequently, these parameters are not ideal for fault detection purpose. Stochastic model based analysis methods were proposed in [5]. In these methods, the feature of multiple AE burst must be extracted first, and then statistical model and clustering criteria are employed to do classification. Inevitably for these methods, large volume of data needs to be processed and baseline criteria need to be established, which make the model less attractive. Gao et al. propose a wavelet transform 1

I.

INTRODUCTION

Currently, vibration is the most widely used tool in diagnosis of machine fault, like bearings, gears, etc. Common vibration sensors include accelerometer, displacement and velocity sensors. Various time and frequency domain analysis techniques have been applied to vibration signal analysis. However, vibration signal have some drawbacks when it come to the incipient of the machine fault. Some early sign of fault in rotation machines might not show in vibration signal but would be caught by AE sensors [1]. If these faults could be

based method to analysis AE signal, which could act as a supplement redundant method for vibration test [6]. He and Li developed a data mining based method to classify the condition indicators derived from different AE data to detect fault [7]. Later, Li and He introduced an EMD-based AE feature quantification method [8]. In their work, successful detection of gear fault is achieved on AE data sampled at as low as 500 kHz. Artificial neural networks (ANN) have also been adopted for AE signal classification [9]. For all of these methods, there are some common short comings. First, the AE signals are sample at several MHz, which make it difficult to process. Second, nearly all of these papers view the AE signal as separate energy burst, but not a consecutive wave form just like any vibration signal. In general, the time domain features extracted from the AE signal are transient time based. These features are then compared with other healthy case reference. They do not give any interpretation of the overall AE signal dynamics along time axis. Take gear fault diagnosis as an example. When a small portion of the gear is damaged or only one tooth has a crack on it, most of energy related AE burst are normal or near normal, only these burst signal generated when the faulty tooth is meshing may have certain abnormal time domain feature. So by statistical analysis in the above reviewed papers, the faulty AE features are not enhanced but weakened, which makes these method have an even more hard time to conclude a fault. A more sound and efficient AE signal processing technique is required. Consider that one actually pre-knows some information about the testing object, for example, the structure of a gear box, the approximate rotation speed of the shaft, etc. In this paper, the authors aim to develop a method which could utilize these kinds of useful information. For vibration signal analysis, methods of interpretation of vibration signals could be classified in three categories: time domain analysis, frequency domain analysis and timefrequency domain analysis. In time domain analysis, besides the features of the waveform itself such as mean, peak, peak to peak, crest factor, root mean square, kurtosis, etc., a more advanced time domain analysis, time synchronous averaging (TSA), is becoming very popular in processing the vibration signal [10-12]. The idea of TSA is to use the ensemble average of the raw signal over a number of revolution in order to get an enhance signal of interest with less noise from other source. For a function x(t), digitized at a sampling interval nT, resulting in sampling in samples x(nT). Denoting the averaged period by mT, TSA is given as [13]: = 1
! !!

order to obtain clean data which only reflects the useful signal of interest, a signal demodulation and frequency reduced technique need to be developed. The remainder of the paper is organized as follow. Section 2 illustrates in details about the methodology. In Section 3, the setup of the experiments for the validation of the methodology is explained. Section 4 discusses the experiment results and illustrates how a fault is identified. Finally, Section 5 concludes the paper. II.
THE METHODOLOGY

A. The Heterodyne Technique First, take a brief look at current AE signal processing steps. The diagram of a commonly taken procedure of AE signal processing is given in Fig. 1.
Acoustic emission sensor (100KHz-1MHz) Pre-amplifier Typically, 20/40/60 dB AD converter Computer post-processing

Figure 1. Traditional AE signal acquisition and preprocessing procedure

In a traditional AE signal processing procedure, all of the data are collected and stored to computer without any advanced signal processing. There are two disadvantages associated with this procedure. First, it makes the data acquisition very expensive. Second, it will highly rely on the computer to process the huge volume of data. When taking a further look at an AE signal, one will find that the AE signal is virtually a carrier signal for the fault signal. The information of interest is related to the load signal, not the high frequency AE carrier signal. In order to get the load signal, demodulation process is required before sampling. This demodulation process is similar to information retrieval in an amplitude/phase modulated radio frequency signal. The carrier signal of a medium wave is radio signal in several MHz, while the information modulated onto that signal is audio signal in kHz. The information is recovered by demodulating the radio signal. After demodulating the carrier using an analog signal conditioning circuit, the acquisition system can then be sampled at audio frequency (10s of kHz). The signal processing can then be performed at lower cost with embedded microcontrollers instead of higher end computers. The AE signal demodulator proposed in this paper would work similarly to a radio frequency quadrature demodulator, effectively shifting the carrier frequency to baseband, followed by low pass filtering. The technique applied here is called heterodyne, which could shift the frequency of the signals to a new frequency range. Mathematically, Heterodyning is based on the trigonometric identity: 1 1 + 2 2 Further, for two signals with frequency f! and respectively, it could be written as = 2 ! 2 ! 1 1 = cos 2 ! ! cos 2 ! + ! 2 2 (2) f! ,


! !!

(1)

More details about TSA could be found in [14]. The successful application of TSA on vibration signal analysis provides the possibility of using it to process AE signal. In order to apply TSA to AE signal analysis, two problems must be addressed. The first problem is to reduce the amount of sampled data while not losing useful information. The second problem is to demodulate the AE signal to recover the machine rotation related frequency. AE signal is modulated by the internal frequency response of the AE sensor itself. In 2

(3)

Where, f! is the carrier frequency of the modulated signal, f! is the demodulators reference input signal frequency. This process could be explained with a simple example. Let f! = 4 Hz and f! = 5 Hz, note y! = sin 2 4 t and y! = sin 2 5 t . Take their multiplication as Y = y! . y! . Their plots are shown below in Fig. 2. Then this multiplication signal could be low pass filtered. The high frequency image at frequency ( f! + f! ) will be removed prior to sampling. This is shown in Fig. 3. In order to successfully shift the signal to a new frequency range, two conditions must be satisfied: (1) Certain nonlinear operations, typically multiplication, must be presented in the heterodyne devices. (2) The frequency components contained in two source signals must be different. Otherwise, only the high frequency signal is left after heterodyning. A detailed discussion of the heterodyne technique applied on the raw AE signal is given in the following. In general, amplitude modulation is the major modulation form for AE signal. Although, frequency modulation and phase modulation could present in the AE signal potentially, they are considered trivial and will not be discussed here. The amplitude modulation function is given in (4). Amplitude modulation function: ! = (! + ) ! (4)
Figure 3. Extraction of the heterodyned signal by low pass filtering Figure 2. The multiplication of two sinusoid signals

Where, U! is the carrier signal amplitude, ! is the carrier signal frequency, m is the modulation coefficient. x is the modulated signal, note as = ! (5) If the modulated signal is not in a sinusoid form, it could be break down into the sum of sinusoid functions in the form of Fourier series. Then, with heterodyne technique, the modulated signal with be multiplied with a unit amplitude reference signal cos (! t). The result is given in the following. ! = ! + ! ! 1 1 = ! + + 2! 2 2 Then substitute (5) in into Eq. (6), 1 1 1 + + ! 2! 2 ! 2 ! 2 1 + ! [ 2! + + 2! 4 ! = (6)

B. AE Signal Processing Using Tach-less TSA TSA is a very useful technique for analyzing vibration data. Basically, two types of TSA algorithm are available in the literature, i.e., TSA with tachometer, and tachometer less TSA. For TSA with tachometer, a tachometer needs to record the shaft speed and the real time angle. But in most applications, the installation of tachometer might be expensive or even impossible. For tachometer less TSA algorithm, although an angular reference signal is still required, it can be derived directly from the tested signal, generally vibration signal, thus no external tachometer is required. The feasibility of extracting a phase reference signal directly from the vibration signal by phase demodulation has been studied in the following papers. Jong et al. proposed a synchronous phase averaging method which could transform a quasi-period signal into a period signal and then perform TSA [15]. They proposed to narrow band filtered out one of the meshing harmonic. The phase variation signal is then calculated by Hilbert transform. Then a standard periodic signal is obtained by non-uniform sampling of the original signal with reference to the phase variation signal. After that, a uniformly sampled periodic signal could be reconstructed by interpolation. Later, standard TSA is calculated and plotted. This was the first paper to show that the gear fault could be identified by discontinuity and irregularity in enhanced TSA waveform. Bonnardot et al. introduced another tachometer less TSA method by angular resampling [11], and later this work was enhanced by Combet and Gelman [12]. They processed 3

(7)

Since U! does not contain any useful information related with the modulated signal, it could be removed by detrend. From (7), it could be seen that only the modulated signal will be preserved after low pass filtering, where the high frequency components around 2! will be removed. The diagram of the proposed frequency down shifting system using heterodyning is shown in Fig. 4. By using the heterodyning technique, the AE data is transformed to the equivalent of vibration data. Two collected examples of faulty AE data and healthy AE data are shown in Fig. 5.

the shocks produced by gear tooth meshing to estimate the position of the gear against time. Then interpolation enables to reconstruct the signal in angular domain. After that the classic TSA method, as shown in (1), could be used to calculate the time synchronous averaging signal.
Gear Box AE Sensor Demodulation board Data Sampling

If the signal average is ideally band pass filtered out one of the meshing harmonics m with a narrow bandwidth W, then the filtered signal could be written by:

Computer

Function Generator

Figure 4. Proposed AE signal acquisition and preprocessing procedure

Despite of the popular application of TSA on analysis of vibration signals, TSA has never been applied to analyze AE signals in the literature. The complicated feature and huge data volume of AE signal make TSA algorithm unrealistic to be performed directly on these data. In this paper, the authors first explore the application of TSA to AE signal analysis. While frequency domain method is performed in the frequency domain, time synchronous average is presented in time domain. This enables the comparison of the acoustic signal produced by each tooth on the same gear as time proceeds. Since many gear failures only involve the deterioration of a small part of the gear, it would be preferred to detect the partially abnormal signal by comparison with the other sound potion. Time synchronous averaging for gear diagnosis generally gives the acoustic signal of a single shaft revolution. If sufficient averages are taken, the time domain averaging signal should approach a standard periodic signal. For healthy gears, the enhanced TSA signal should display a fully or near fully periodic characteristic. While for the faulty gear, it is assumed there will be some abnormal behavior in the TSA signal. This judgment will be verified in the following part of this paper. Consider an AE signal of a gear box, adopted from vibration signal [14], written as:
!

Figure 5. Examples of faulty and healthy AE data collected with heterodyne

! ! [1 + ! ]cos [2 + ! + ! ]

(10)

=
! !!

! (2 ! + ! ) +

(8)

Notice that in (10), all of the modulation information remains intact. Any fault related modulation effect on the signal is preserved. For some simple vibration signal, when the amplitude and phase modulation function is obtained, by inspecting the modulation function, the mechanical fault could be detected [16]. But for AE signal, the modulation process is complicated and contaminated, it is not easily to determine whether there is fault or not by simply looking at the modulation function. It is then needed to go one step further to calculate the TSA of the signal. Since no real time tachometer synchronized with the data acquisition process is available. A phase reference signal needs to be derived from the AE signal. In (10), z! (t) could be viewed as the real part of a complex function ! (t) [16], known as the analytic signal and defined by ! = ! [! ] Where H(z! t ) is the Hilbert transform of z! t . Then the instantaneous phase could be calculated as = ( [! ]/! ) (12) The instantaneous frequency (IF) is defined as the first derivative of the instantaneous phase as a function of time = ( ) 4 (13) (11)

Where, X ! is the mth order harmonics, N tooth number, f shaft frequency, ! the initial phase angle. Suppose this signal is amplitude and phase modulated by the function ! and ! , respectively. An important point need to mention is that, any fault in the gear will impose extra modulation effect on the signal. Here, for the simplicity of notation, it is included in ! and ! . Then the amplitude and phase modulated signal could then be expressed as:
!

=
! !!

! [1 + ! ] cos [2 + ! + ! ] +

(9)

There are several ways to calculate the instantaneous frequency [17]. In this work, the derivative of the unwrapped phase signal is taken. The zero crossing time is then computed from the instantaneous frequency. A more detailed explanation of extract phase reference is provided in [11][17]. In order to extract a phase reference signal, it is needed to firstly perform Hilbert transform on the narrow band filtered signal and then construct the analytic signal and calculate the instantaneous frequency. To determine the narrow band filter range, the meshing frequency is needed. Although according to the shaft speed, the meshing frequency could be calculated, however, a more accurate meshing frequency is obtained from the power spectrum density (PSD). The PSD corresponding to the data shown in Fig.5 is given below in Fig. 6.

us to inspect a single harmonic to detect the hidden fault. By TSA the irregular modulation effect will be enhanced and easily observed. The reason why the demodulated AE signal need to be narrow band filtered is explained as follows: The demodulation process could only demodulate the AE signal modulated on a specific carrier frequency. The signal modulated on other frequency bands will be shifted down to a frequency band around their base bands and will act as noise. In order to enhance the signal to noise ratio, the narrow band pass filter is proposed. Classic TSA is also tested without band pass filter, the fault feature is not revealed as expected, which then lead to this improved TSA methods. III. EXPERIMENTAL TEST RIG CONFIGURATION In this section, the experiment to verify the proposed AE signal processing method is presented. In Fig. 7, the demodulation board (Analog devices - AD8339) and sampling devices (NI-DAQ 6211) are shown. These devices were provided by NRG Systems, Inc. The demodulation board performed the multiplication of sensor signals and reference signals. It is an analog device and much more affordable than a high rate sampling board. It takes two inputs, one from the AE sensor, and another from function generator as reference signal. The basic principle of AD8339 could be explained by Gilbert cell mixers. In electronics, the Gilbert cell is commonly used as an analog multiplier and frequency mixer. This circuits output current is an accurate multiplication of the base currents of both inputs. According to (3), it could convert the signal to baseband and twice the carrier frequency. This demodulation board could support up to 4 channels with analog inputs as high as 40MHz. These features make it well satisfy AE signal demodulation requirement. The output of the demodulation board goes to the sampling board and the high frequency component is filtered out. NI-DAQ 6211 is a low frequency data acquisition device, with a sample frequency up to 250kS/s. Before data acquisition, one needs first to determine the frequency of the reference signal for demodulation. The purpose is to down shift the AE signal frequency as low as possible. In order to remove the carrier frequency, the reference signal frequency needs to be equal to or close to the AE carrier frequency. This is the most challenging part of utilizing the heterodyne technique. Note that raw AE signal is a wide band signal with complicated modulation. The purpose here is to demodulate the component with highest amplitude. Each AE sensor has its specific frequency response range, which further depends on the testing system they are mounted on. With reference to the AE sensor user manual, a coarse range of the sensor response frequency is given. In order to identify a more accurate AE sensor response frequency, it is proposed to use a function generator with sweep function to test the system and record the output. With a wide range of sweep frequency signal as the reference signal to demodulation board, the demodulation result varies accordingly. As mentioned before, the energy impact information is modulated on high frequency carrier signals in a wide frequency band. When the strongest carrier signal is successfully demodulated, the highest amplitude will display in the output signal. This means the demodulation frequency matches the strongest 5

Figure 6. Power spectrum of faulty and healthy data after Heterodyne

According to [15], a periodic signal can be reconstructed from the original signal by non-uniform sampling and time synchronous averaging. A similar procedure is followed in this work. However, raw AE signal can not be transformed into periodic signal because AE signal has a lot of complicated frequency components. Hence, instead of performing TSA on the raw AE data, a narrow band filtering is proposed to filter out the meshing harmonics and the filtered AE data will be used to perform TSA. In this paper, a zero-phase filter with a bandwidth of only 2Hz is designed. Then, through classic TSA technique, an enhanced periodic signal of a single meshing harmonic frequency could be calculated. The procedure of the modified TSA algorithm for processing AE data is given below. The Modified TSA Procedure is listed below: Calculate the power spectrum density. Identify the base meshing frequency and the harmonics frequency. Construct the analytic signal based on the meshing frequency and the raw AE data. Narrow band filter the analytic signal and take the unwrapped version of the phase, and then the angular position of the shaft could be extracted as proposed in [11]. Narrow band filter the original data. Notice that zero-phase filtered is required in order to prevent phase shifting of the signal. Perform TSA algorithm on the filtered data. Inspect the wave form of TSA signal for any irregular behavior to identify gear fault.

This modified TSA algorithm is performed on a zero-phase filtered harmonic signal instead of on the original signal. As shown in (10), the fault related modulation function is preserved in the meshing harmonics after filtering. It enables

modulation frequency of the AE sensor. Otherwise, the energy related information will still dwell in high frequency end and therefore would be lost after low pass filtering and sampling. With sweep frequency input, one can analyze the energy level of the signal at different demodulation frequency. The AE sensor modulation frequency would be identified when the demodulated AE signal have the largest energy level. By this method, a more accurate frequency range of the sensor output could be found. In the performed experiment, it was identified 400kHz as the major AE carrier signal frequency. This frequency is used as the demodulation reference frequency.

As for signal acquisition, Labview signal express software was used. The data sampling rate of the acquisition could be modified in setting options. Even if the data sample board could support up to 250kHz sampling rate, it was set to only 100 kHz for the tests.

Figure 9. AE sensor location on test rig

Figure 7. Demodulation device and data sampling board

The proposed method was then tested on a two stage split torque gearbox. Compared with traditional gear box, the split torque gear box has an intermediate stage which could split the torque and change the transmission ratio between input and output shaft. The structure of the gear box is given in Fig. 8. Since this is speed reduction gear box, the input side and the output side have a 2.4 times reduction ratio. For the faulty split torque gear box, at the output side, one of the intermediate gears with 48 teeth had a 50% tooth loss on one of the teeth as shown in Fig.8. The location of the sensor on the gearbox is shown in Fig. 9. The AE sensor was located on the gear house close to the faulty gear. Note that this paper is primarily to evaluate the heterodyne technique and the feasibility of time synchronous averaging on AE data, so a relatively large fault is created on the gear. Incipient fault like initial crack and tooth surface wear will be tested in the future.

In order to verify the effectiveness of the proposed method, both healthy and faulty gear box were tested with input shaft speeds: 10Hz, 20Hz, 30Hz, 40Hz, 50Hz, and 60Hz. In the whole experiment process, the gear box has 0 loading. With loadings the faulty gear feature will be enlarged due to larger impact on each tooth, and therefore the faulty feature in AE could be more easily detected. In this 0 load experimental setting, the identification of gear fault is more difficult than loaded cases. As mention above, this gear box has a total of 2.4 time speed reduction ratio. In table 1, the output shaft frequency and the interested shaft (faulty gear shaft) frequency are given under the tested speed.
TABLE 1. OUTPUT SHAFT SPEED CORRESPONDING TO INPUT SHAFT SPEED Input shaft speed (Hz) Faulty gear shaft frequency (Hz) Output shaft speed (Hz) 10 5.56 4.17 20 11.1 8.33 30 16.7 12.5 40 22.2 16.7 50 27.8 20.8 60 33.3 25

IV.

THE RESULTS

In order to get a consistent result, the base band meshing signal is chosen to do TSA for all of the following result. But the method works as well with the meshing harmonics. Results show that all 10Hz-60Hz faulty TSA signals show an obvious wave form defect. While there is faulty feature involved, some of the teeth are not meshing normally. This will affect the wave shape in some cycles as well the overall dynamic of the shaft. Therefore, the faulty signal will have some underdeveloped wave cycles. Now, the analysis results of the experiments on the test-rig are presented in Fig. 10 through Fig. 15. Each figure shows the result under one operation speed. Both the fault TSA result and the healthy TSA result are shown for comparison.
Figure 8. Dynamic model of the experimental split torque gearbox

Figure 10. Faulty and healthy TSA at 10Hz input speed over one rev.

From Fig. 10 Fig. 15, it is easy to observe the abnormal meshing behavior corresponding to the faulty tooth. A waveform detect, in the form of undeveloped cycles or cycles missing, could be seen from all of the faulty TSA data. Intuitively, since there was a 50% tooth loss, the faulty tooth would not mesh normally as other health teeth on the gear. From the perspective of AE signal, the meshing impact corresponding to that specifically cycle is corrupted. Thus, waveform defect will be presented. While in the health case, the wave form of the TSA signal is almost pure periodic with each cycle fully developed.

Figure 12. Faulty and healthy TSA at 30Hz input speed over one rev.

Figure 13. Faulty and healthy TSA at 40Hz input speed over one rev.

Figure 11. Faulty and healthy TSA at 20Hz input speed over one rev.

V.

CONCLUSIONS

Figure 14. Faulty and healthy TSA at 50Hz input speed over one rev.

In this paper, a new methodology for gear fault diagnosis suing AE sensors is presented. In comparison with existing AE signal processing techniques, this technique gives an insight interpretation of the internal physical characteristics of the gearbox and AE signals. Previously, researchers treated AE signals as separate energy bursts, focusing mainly on the characteristic within the burst, but ignored the time continuous features, which is closely related to the faults in a gearbox. By comparison of the signal features along the time axis, abnormal behavior occurred repeatedly in a rotating machine could be extracted and detected. Specifically, heterodyning technique is utilized to demodulate the AE sensor signals at the first stage. Using the heterodyning demodulator, the AE signal frequency can be down shifted to below 50kHz with most of the high frequency carrier signal removed. After that, a modified TSA based approach can be used to process the AE data. In the presented method, the TSA signals of the meshing harmonics instead of the raw signals are used. By inspecting the TSA signal waveform defect, a tooth fault can be detected. The presented method could be utilized on most rotational machine fault diagnosis when AE sensor is used. The system cost could be largely reduced in comparison with traditional methods. This paper has established a foundation that AE signals could be sampled at a relatively low rate while not losing any useful information significantly. It also makes the low cost industrial application of AE based fault detection feasible in practice. REFERENCES
[1] A.M. Al-Ghamd and D. Mba, A comparative experimental study on the use of acoustic emission and vibration analysis for bearing defect identification and estimation of defect size, Mechanical Systems and Signal Processing, Vol. 20, No. 7, pp. 1537 1571, 2006. [2] T. Holroyd, Acoustic emission & ultrasonics monitoring handbook, Oxford, U.K.: Coxmoor Publishing Co., 2000. [3] J. Shiroishi, Y. Li, S. Liang, T. Kurfess and S. Danyluk, Bearing condition diagnostics via vibration and acoustic emission measurements, Mechanical Systems and Signal Processing, Vol. 11, No. 5, pp. 693 - 705, 1997. [4] D. Mba, Acoustic emissions and monitoring bearing health, Tribology Transactions, Vol.46, No.3, pp. 447 - 451, 2003. [5] L. D. Hall and D. Mba, Acoustic emissions diagnosis of rotor-stator rubs using the ks statistic, Mechanical Systems and Signal Processing, Vol. 18, No. 4, pp. 849 - 868, 2004. [6] L. Gao, F. Zai, S. Su, H. Wang, P. Chen and L. Liu, Study and application of acoustic emission testing in fault diagnosis of low-speed heavy-duty gears, Sensor, Vol.11, No.1, pp. 599 611, 2011. [7] D. He, R. Li, J. Zhu and M. Zade, Data mining based full ceramic bearing fault diagnostic system using ae sensors, IEEE Transactions on Neural Network, Vol. 22, No. 12, pp. 2022 - 2031, 2011. [8] R.Li and D. He,Rotational machine health monitoring and fault detection using emd-based acoustic emission feature quantification, IEEE Transactions on Instrumentation and Measurement , Vol. 61, No.4 , pp. 990 - 1001, 2012. [9] J.S. Kwak and M.K. Ha, Neural network approach for diagnosis of grinding operation by acoustic emission and power signals, Journal of Materials Processing Technology , Vol.147, No.1, pp. 65 - 71, 2004. [10] P.D. McFadden, M.M. Toozhy, Application of synchronous averaging to vibration monitoring of rolling element bearings, Mechanical Systems and Signal Processing, Vol. 14, No. 6, pp.891906, 2000.

Figure 15. Faulty and healthy TSA at 60Hz input speed over one rev.

[11] F. Bonnardot, M. El Badaoui, R.B. Randall, J. Daniere, and F. Guillet, Use of the acceleration signal of a gearbox in order to perform angular resampling (with limited speed fluctuation) , Mechanical Systems and Signal Processing, Vol. 19, No. 4, pp. 766 785, 2005. [12] F. Combet and L. Gelman, An automated methodology for performing time synchronous averaging of a gearbox signal without speed sensor, Mechanical Systems and Signal Processing, Vol. 21, No. 6, pp. 2590 2606, 2007. [13] S. Braun, The extraction of periodic waveforms by time domain averaging. Acustica, Vol 32, No.2, pp. 6977, 1975. [14] P.D. McFadden, A revised model for the extraction of periodic waveforms by time domain averaging, Mechanical Systems and Signal Processing, Vol.1, No.1, pp. 83 - 95, 1987. [15] J. Jong, J. McBride, J. Jones, T. Fiorucci and T. Zoladz, Synchronous phase averaging method for machinery diagnostics, Defense Technical Information Center, Proceedings of a Joint Conference, Mobile, Alabama, April 22-26, 1996. [16] P.D. McFadden, Detecting fatigue cracks in gears by amplitude and phase demodulation of the meshing vibration, Journal of Vibration, Acoustics, Stress, and Reliability in Design, Vol. 108, No.2, pp. 165 170, 1986. [17] M. Feldman, Hilbert transform in vibration analysis, Mechanical Systems and Signal Processing, Vol. 25, No.3, pp. 735 - 802, 2011.

You might also like