Mohd Sufian OTHMAN1, Mohd Zaki NUAWI1, Ramizi MOHAMED2

Universiti Kebangsaan Malaysia, Dept. of Mechanical & Material Eng. (1), Dept. of Electrical, Electronic & Systems Eng. (2)

doi:10.15199/48.2016.11.51

Experimental Comparison of Vibration and Acoustic Emission

Signal Analysis Using Kurtosis-Based Methods for Induction
Motor Bearing Condition Monitoring
Abstract. Vibration and acoustic emission (AE) signal measurement are established tools in condition monitoring of bearings. This paper presents a
comparison between two statistical methods, i.e. kurtosis, and another kurtosis based method called I-kaz using simultaneous signal of vibration and
AE. The existence of harmonics fault frequencies in envelope spectrums also have been discussed. The results reveal that both signals are suitable
for induction motor bearing fault detection.

Streszczenie. W artykule przedstawiono porównanie dwóch metod statystycznych, kurtozy oraz I-kaz używająnych do diagnostyki łożyskowania
maszyn elektrycznych. Jako sygnały diagnostyczne wykorzystuuje się pomiar wibracji i emisji akustycznej.(Eksperymentalne porównanie drgań i
analizy sygnałów emisji akustycznej do monitorowania stanu łożysk).

Keywords: bearing fault, kurtosis, I-kaz.

Słowa kluczowe: monitorowanie stanu łożyska, kurtoza, I-kaz.

1. Introduction 3
1 N
Bearing is one of important components in rotating
machines that enable smooth acceleration and reduce N
  yn   
friction. Ball and roller bearing are common type of bearing
(3) Sk  n 1 .
s3
used in induction motors. Normally, an induction motor has 4
1 N
two bearings supporting the rotor shaft; at the fan-end side,
and at the drive-end side. The faulty bearing can cause N
 y n  
major defects to motor if not fixed in a timely manner. The (4) Kur  n 1 .
common failure observed in bearing elements are inner s4
race fault and outer race fault. where s, is the square root of the variance, or the standard
Vibration analysis is the most widely used technique for deviation as stated in Eq. 5.
fault detection of rotating machineries and components (5) s  .
such as bearings and gears. As vibration can be The usage of kurtosis for bearing fault detection was
represented by the displacement, velocity or acceleration, first proposed by Dyer and Stewart in 1978 [8]. For a good
any sensor that measure any of these variables can be condition bearing with normal or Gaussian distribution, the
used. Piezoelectric accelerometer is the most popular kurtosis value is close to 3. However, when the fault is well
detector used to measure vibration because of rugged, advanced, the kurtosis value was reported to come down to
small and relatively inexpensive [1]. Therefore, this study the same value as undamaged bearing which is 3.
focused on acceleration measurement using an Therefore, to solve this issue they have suggested to
accelerometer. Acoustic emission (AE) is the phenomena of measure kurtosis in selected frequency bands.
transient elastic wave generation as a result of fast strain Furthermore, kurtosis also have been applied in Spectral
energy discharge from deformation or localised damage Kurtosis analysis [9], [10], or also known as kurtogram
within or on material surface [2], [3]. Combination of method [11], [12] for detection of bearing fault in frequency
vibration and AE measurement in bearing diagnostics has domain. Another great kurtosis-based method is the
been done by Al-Ghamd and Mba [4], Eftekharnejad et al. integrated kurtosis-based algorithm for z-filter technique (I-
[5], and Liu et al [6] using various methods. kaz) proposed by Nuawi et al [13]. This method provides an
Statistical analysis is one of the earliest approaches for I-kaz coefficient and a three dimensional graph with low,
bearing fault detection in time-domain. The calculation of high, and very high frequency range which are represented
crest factor, which is the ratio of peak value to the root in x-axis, y-axis, and z-axis respectively. Detail of signal
mean square (RMS) value of signal data have been the decomposition process is as follows:
simplest approach to detect bearing faults in time domain
analysis [7]. However, this method had limited success in  Low frequency range (L): 0 to 0.25 fmax
the detection of localized defects. Generally, most of  High frequency range (H): 0.25 to 0.5 fmax
parameters used in the time domain are derived from  Very high frequency range (V): 0.5 fmax to fmax
statistical moment of data. The first and second moments
are well known, being the mean and the variance where fmax is the half value of the data sampling rate. I-kaz
respectively. For N numbers of data set (y1,y2,…,yn), the coefficient, Z is derived from the kurtosis and standard
mean,  and the variance,  can be calculated as: deviation from all frequency ranges, as in Eq. 6.
1 N 1
 y  .
4 4 4
(1)  (6) Z   KurL  sL  KurH  sH  KurV  sV .
n
N n 1
N
2 The I-kaz method have been successfully applied for
1 N
(2)  
N
  yn   
n 1
. condition monitoring of car engine bearing [14], engine
block [15], suspension system [16], machine cutting tool
The third and fourth moments, the skewness (Sk) and the [17]–[19], and charpy impact tester [20].
kurtosis (Kur) respectively, defined as

208 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 92 NR 11/2016

 This paper presents a comparison of vibration and AE Table 1: Motor Specification (Nameplate)
signal analysis using three fault detection techniques, i.e. Make Leybold Didactic GmbH
(1) time-domain waveform analysis using kurtosis method Model Number 732 94
and envelope spectrum analysis to locate fault frequencies, Rated Power 0.7 kW
-1
(2) I-kaz graphical and coefficient methods, and (3) Rated Speed 1405 min
Frequency 50 Hz
comparative study of I-kaz coefficient against kurtosis.
Starting Current 2.0 A
2. Methodology
The proposed work flow chart for this research is shown Three bearing condition have been tested using the
in Fig. 1. The process was started with an experimental test same motor which are normal bearing, bearing with inner
rig setup followed by data acquisition with two types of race fault, and bearing with outer race fault. All faulty
sensors: (a) accelerometer for measuring vibration signal, bearings have been artificially damaged by axial drilled of a
and (b) AE transducer for acoustic emission signal. Data 1mm hole through the raceways (see Fig. 3).
were collected and stored in the computer before analysis (a) (b)
using MATLAB version 8.3 have been done.

Start

Test rig setup

Normal bearing Outer-race fault Inner-race fault

Data acquisition
Fig. 3: Tested bearing with artificial localized defect at (a) inner
race and (b) outer race.
Vibration signal AE signal
The motor has been tested with different tork applied
started with 0 Nm and increased 1 Nm until reach full load
(3 Nm). The motor shaft speed were measured and
Waveform and spectrum analysis
recorded as tabulated in Table 2.
I-kaz analysis Table 2: Motor shaft speeds at different tork.
Tork 0 Nm 1 Nm 2 Nm 3 Nm
Comparative analysis Speed 1450 rpm 1430 rpm 1420 rpm 1380 rpm

Tested bearing was placed on the drive-end of the

Stop motor. Detail information about the tested bearing is stated
in Table 3. Vibration signal was measured using Endevco
Fig. 1: Proposed research flowchart 751-100 piezoelectric accelerometer with sensitivity of
100mV/g and a DeCI SE1000H acoustic emission
The experimental tests were performed on a test rig
transducer with sensitivity of 100V/µm used to collect AE
consist of a low speed 4 pole induction motor with the rated
signal (Fig. 4).
speed of 1405 rpm (See Table 1). The motor was supplied
Data were sampled at 99,968 Hz using PicoScope
by the 3 phase variable voltage transformer pre-set at
6402A oscilloscope at one second time frame and stored in
380V. Tacho generator was connected to the drive-end of
32 files. Hence, the total acquired are 3,198,176 data.
motor to measure the rotation speed of the induction motor.
Collected data for both signals were in voltage (V). For the
A magnetic powder brake was coupled to the other end of
vibration signal, the data have been converted to
tacho generator. This magnetic brake used to apply axial -2
acceleration (g/ms ) to represent vibration by dividing the
loads to the motor, controlled by the control unit. The details
voltage to sensor sensitivity value.
of test rig setup are shown in Fig. 2.
Table 3: Testing bearing information
(1) (2) (3)
Bearing Number FAG 6204ZR C3
Bearing outer diameter 47mm
Bearing inner diameter 20mm
Bearing width 14mm
Ball diameter 7.938mm
Number of balls 8
Contact angle 0
Table 4: Tested Bearing Fault Frequencies at Different Shaft
Speeds in Hertz (Hz).
Frequency Name 1380 rpm 1450 rpm
(4) (5) (6) (7) (8) Shaft frequency, fS 23.0 24.2
Inner race fault (BPFI), fIR 113.8 119.6
Fig. 2: Experimental setup: (1) computer, (2) oscilloscope, (3) Outer race fault (BPFO), fOR 70.2 73.8
control unit, (4) 3 phase variable voltage supply, (5) motor
protection switch, (6) squirrel cage induction motor, (7) tacho Table 4 is presented the tested bearing fault frequencies
generator, and (8) magnetic powder brake. in four different shaft speed. In this table, 1380 rpm is for full
load and 1450 rpm is for no load. These fault frequencies
were used in the envelope spectrum analysis in frequency-
domain plots.

PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 92 NR 11/2016 209

 the AE signal. In no load plots, the kurtosis of normal
bearing is close to 3 for both acceleration and AE signal,
proved that the data are normally distributed [Refer Fig. 5(a)
and Fig. 5(b)]. Kurtosis values for faulty bearings are
significantly higher than 3 as shown in Fig. 5(c) to Fig. 5(f).

(a) (b)

(a) (b)

(c) (d)

Test bearing location

Fig. 4: Sensor location on motor casing: (a) Piezoelectric

accelerometer, and (b) AE transducer.

3. Results and Discussion

Three analyses have been carried out; i.e. (a) waveform
and spectrum comparison, (b) I-kaz analysis, and (c) (e) (f)
comparative analysis. The results show in the following sub-
section were taken from the first file out of 32 except for
sub-section 3.3. which was a combination of all data.
3.1 Time Waveform & Envelope Spectrum Analysis
In the time waveform analysis, kurtosis value of
acceleration and AE signal have been compared between
normal and faulty bearing. The results were grouped and Fig. 6: Comparison of time waveform and envelope spectrum for
presented based on motor speeds and loads. On the other acceleration and AE signal of the motor running at 1380 rpm (full
hand, fundamental and harmonic fault frequencies in load) with (a),(b) normal, (c),(d) inner race faulty, and (e),(f) outer
envelope spectrum plots also have been observed and race faulty bearing.
reported accordingly.
Same notification also observed in full load except for
(a) (b) the kurtosis value of AE signal in the inner race faulty
bearing which was considered small compared to
acceleration signal on the left side [see Fig. 6(c) and Fig.
6(d)]. In both no load and full load envelope analysis, the
outer race fault frequency and it repetitions are clearly
observed for both acceleration and AE signal [See Fig. 5
(e),(f) and Fig. 6(e),(f)]. However, for the inner race fault,
the repetition frequencies are not discovered, although the
fault frequency appeared perfectly [See Fig. 5(c),(d) and
(c) (d)
Fig. 6(c),(d)].
3.2 I-kaz Analysis
The shape and the scattering of graphical I-kaz plots for
acceleration and AE signal were observed and compared
between normal and faulty bearing to check for any
discrepancy. In addition, the I-kaz coefficient (Z) for each
plots also have been analysed. Axis ranges for all plots in
(e) (f) each figure have been standardised for easy and equal
comparison.
Results for I-kaz analysis are displayed in Fig. 7 for no
load and Fig. 8 for full load. It is observed that all graphical
I-kaz plots provided clear identification to verify bearing
conditions accordingly. For normal bearing, both
acceleration and AE signal data are concentrated at same
slot for all low, high and very high frequency ranges with Z
for acceleration signal is quite similar for no load and full
Fig. 5: Comparison of time waveform and envelope spectrum
for acceleration and AE signal of the motor running at 1450 rpm (no load plots whereas Z for AE signal for no load is slightly
load) with (a),(b) normal, (c),(d) inner race faulty, and (e),(f) outer smaller than full load plot. Different finding is observed for
race faulty bearing. faulty bearings where Z for acceleration data of inner race
and outer race faults in full load were larger than in no load,
Time waveform and envelope spectrum of vibration and but Z for AE data in full load was smaller than in no load
AE signal are shown in Fig. 5 and Fig. 6, represented no plots. However, when comparing separately in the same
load and full load respectively. In all these figures, (a), (c), figure, the I-kaz coefficient values for acceleration and AE
and (e) are plots for acceleration and (b), (d), and (f) are for signal in both no load (Fig. 7) and full load (Fig. 8) shown

210 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 92 NR 11/2016

significant increment between normal and damaged plots for vibration and AE signal with values grouped by
bearings. For example, the Z for acceleration signal of bearing conditions, i.e. normal, inner race faulty, and outer
normal bearing is 4.286×10-9 as in Fig. 7(a), 3.77×10-8 for race faulty bearing have been compared visually. Fig. 9 and
inner race fault as in Fig. 7(c), and 3.348×10-7 for outer race Fig. 10 shown scatter plot for the without load and with full
fault as in Fig. 7(e). load respectively.
(a) (b) (a) (b)

(c) (d)
Fig. 9. Scatter plot of I-kaz coefficient vs Kurtosis for the motor
running at 1450 rpm (no load).

(a) (b)

(e) (f)

Fig. 10. Scatter plots of I-kaz coefficient vs Kurtosis for the 4 pole
motor running at 1380 rpm (full load).

In all figures, normal bearing plots show consistent

Fig. 7. Comparison of I-kaz graphic and coefficient for position compared to inner race and outer race faults with
acceleration and AE signal of the motor running at 1450 rpm (no the vibration signal provided better stability compared to AE
load) with (a),(b) normal, (c),(d) inner race faulty, and (e),(f) outer signal. Plots of acceleration and AE signal for inner race
race faulty bearing. and outer race faulty bearing are scattered at different slots.

(a) (b) 4. Conclusion

From overall experimentation results, vibration and AE
signals analysis using kurtosis based methods have
provided good identification to distinguish between normal
and damaged bearings. On the other hand, the graphical I-
kaz plot also can be quick and easy technique to detect
bearing faults. Furthermore, the comparison of I-kaz
coefficient against kurtosis values from all data files using
(c) (d)
the scatter plot given better clarification.

Authors: Ir., Mohd Sufian Othman, Dept. of Mechanical and

Material Engineering, Universiti Kebangsaan Malaysia, 43600
Bangi, Selangor, Malaysia. Email: [email protected];
Dr., associate prof, Mohd Zaki Nuawi, Dept. of Mechanical and
Material Engineering, Universiti Kebangsaan Malaysia, 43600
Bangi, Selangor, Malaysia. Email: [email protected];
Dr., Ramizi Mohamed, Dept of Electrical, Electronic and System
Engineering, Universiti Kebangsaan Malaysia, 43600 Bangi,
(e) (f) Selangor, Malaysia. EmaIl: [email protected].

REFERENCES
[1] R. Bogue, “Sensors for condition monitoring: a review of
technologies and applications,” Sens. Rev., vol. 33, no. 4, pp.
295–299, 2013.
[2] Y. He, X. Zhang, and M. I. Friswell, “Defect Diagnosis for
Rolling Element Bearings Using Acoustic Emission,” Journal
of Vibration and Acoustics, vol. 131, no. 6. p. 061012, 2009.
[3] S. Al-Dossary, R. Hamzah, and D. Mba, “Observations of
Fig. 8. Comparison of I-kaz graphic and coefficient for changes in acoustic emission waveform for varying seeded
acceleration and AE signal of the motor running at 1380 rpm (full defect sizes in a rolling element bearing,” Appl. Acoust., vol.
load) with (a),(b) normal, (c),(d) inner race faulty, and (e),(f) outer 70, no. 1, pp. 58–81, 2009.
race faulty bearing. [4] A. M. Al-Ghamd and D. Mba, “A comparative experimental
study on the use of acoustic emission and vibration analysis
3.3 Comparative Analysis for bearing defect identification and estimation of defect size,”
Comparative analysis between I-kaz and kurtosis from Mech. Syst. Signal Process., vol. 20, no. 7, pp. 1537–1571,
all recorded data (32 files) have been performed. Scatter 2006.

PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 92 NR 11/2016 211

[5] B. Eftekharnejad, M. R. Carrasco, B. Charnley, and D. Mba, Kurtosis-based Algorithm for Z-filter technique,” J. Appl. Sci.,
“The application of spectral kurtosis on Acoustic Emission vol. 8, no. 8, pp. 1541–1547, 2008.
and vibrations from a defective bearing,” Mech. Syst. Signal [14] N. I. I. Mansor, M. J. Ghazali, M. Z. Nuawi, and S. E. M.
Process., vol. 25, no. 1, pp. 266–284, 2011. Kamal, “Monitoring bearing condition using airborne sound,”
[6] X. Liu, X. Wu, and C. Liu, “A comparison of acoustic emission Int. J. Mech. Mater. Eng., vol. 4, no. 2, pp. 152–155, 2009.
and vibration on bearing fault detection,” in 2011 International [15] M. Z. Nuawi, S. Abdullah, A. R. Ismail, R. Zulkifli, M. K.
Conference on Transportation, Mechanical, and Electrical Zakaria, and M. F. H. Hussin, “A Study on Ultrasonic Signals
Engineering (TMEE), 2011, pp. 922–926. Processing Generated From Automobile Engine Block Using
[7] N. Tandon and A. Choudhury, “A review of vibration and Statistical Analysis,” WSEAS Trans. Signal Process., vol. 4,
acoustic measurement methods for the detection of defects no. 5, pp. 279–288, 2008.
in rolling element bearings,” Tribol. Int., vol. 32, no. 1999, pp. [16] S. Abdullah, M. Z. Nuawi, M. Z. Nopiah, and A. Ariffin, “Study
469–480, 2000. on correlation between strain and vibration signal using
[8] D. Dyer and R. Stewart, “Detection of Rolling Element hybrid I-Kaz method,” Contin. Mech. Fluids, Heat, vol. 6, no.
Bearing Damage by Statistical Vibration Analysis,” J. Mech. 3, pp. 79–88, 2010.
Des., vol. 100, no. 2, pp. 229–235, 1978. [17] M. Z. Nuawi, F. Lamin, A. R. Ismail, S. Abdullah, and Z.
[9] W. Guo, P. W. Tse, and A. Djordjevich, “Faulty bearing signal Wahid, “A Novel Machining Signal Filtering Technique: Z-
recovery from large noise using a hybrid method based on notch Filter,” World Acad. Sci. Eng. Technol. 54, vol. 3, no. 6,
spectral kurtosis and ensemble empirical mode pp. 24–29, 2009.
decomposition,” Meas. J. Int. Meas. Confed., vol. 45, no. 5, [18] M. A. F. Ahmad, M. Z. Nuawi, S. Abdullah, Z. Wahid, Z.
pp. 1308–1322, 2012. Karim, and M. Dirhamsyah, “Development of Tool Wear
[10] V. C. M. N. Leite, J. Guedes, G. Francimeire, C. Veloso, L. Machining Monitoring Using Novel Statistical Analysis
TM
Eduardo, S. Member, G. Lambert-torres, E. L. Bonaldi, L. Ely, Method, I-kaz ,” Procedia Eng., vol. 101, pp. 355–362,
and D. L. De Oliveira, “Detection of Localized Bearing Faults 2015.
in Induction Machines by Spectral Kurtosis and Envelope [19] J. A. Ghani, M. Rizal, A. Sayuti, M. Z. Nuawi, M. N. Ab.
Analysis of Stator Current,” IEEE Trans. Ind. Electron., vol. Rahman, and C. H. Che Haron, “New Regression Model and
62, no. 3, pp. 1855–1865, 2015. I-Kaz Method for Online Cutting Tool Wear Monitoring,”
[11] X. Zhang, J. Kang, L. Xiao, J. Zhao, and H. Teng, “A New World Acad. Sci. Eng. Technol. 60, no. 2008, pp. 420–425,
Improved Kurtogram and Its Application to Bearing Fault 2009.
Diagnosis,” Shock Vib., vol. 2015, p. 22 pages, 2015. [20] M. B. Ali, S. Abdullah, M. Z. Nuawi, M. M. Padzi, and K. A.
[12] Y. Lei, J. Lin, Z. He, and Y. Zi, “Application of an improved Zakaria, “Experimental Analysis of an Instrumented Charpy
kurtogram method for fault diagnosis of rolling element Impact Using Statistical Study Based Data Analysis,” Int. J.
bearings,” Mech. Syst. Signal Process., vol. 25, no. 5, pp. Mech. Mater. Eng., vol. 6, no. 2, pp. 260–268, 2011.
1738–1749, Jul. 2011.
[13] M. Z. Nuawi, M. J. M. Nor, N. Jamaludin, S. Abdullah, F.
Lamin, and C. K. E. Nizwan, “Development of integrated

212 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 92 NR 11/2016