Compound Faults Detection in Gearbox Via Meshing Resonance and Spectral Kurtosis Methods.2016.12.041

Journal of Sound and Vibration ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Contents lists available at ScienceDirect
Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi
Compound faults detection in gearbox via meshing resonance

and spectral kurtosis methods
Tianyang Wang, Fulei Chu n, Qinkai Han, Yun Kong
Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China
a r t i c l e i n f o abstract
Article history: Kurtosis-based impulsive component identification is one of the most effective algorithms
Received 18 September 2016 in detecting localized faults in both gearboxes and rolling bearings. However, if localized
Received in revised form faults exist in both gear tooth and rolling bearing simultaneously it is difficult to tell the
5 December 2016
differences between the two types of defects. As such, this study proposes a new method
Accepted 24 December 2016
to solve the problem by using the meshing resonance and spectral kurtosis (SK) algo-
rithms together. In specific, the raw signal is first decomposed into different frequency
Keywords: bands and levels, and then the corresponding Kurtogram and MRgram are calculated via
Compound faults the fault SK analysis and the meshing index. Furthermore, the resonance frequency bands
Gearbox
induced by localized faults of the gear tooth and rolling bearing are separately identified
Rolling bearing
by comparing the Kurtogram and the MRgram. Finally, the compound faults are respec-
Meshing resonance
Spectral kurtosis tively detected using envelope analysis. The effectiveness of the proposed method has
MRgram been validated via both simulated and experimental gearboxes vibration signals with
Kurtogram compound faults.
Fault diagnosis & 2017 Elsevier Ltd All rights reserved.
1. Introduction
Localized faults appear in the main components of machinery because of severe and long-term working conditions. For
the rotating machinery, localized defects always result in a series of impulses with certain repeated frequencies, such as the
ball pass frequency of the rolling bearing, and the rotational frequency of the faulty gear, which can be used as fault
characteristic frequencies to realize further diagnosis of faults. In fact, detecting localized faults in the rolling bearings and
gearboxes has always been an important topic in both academia and industry [1,2]. The condition of rotating machinery can
be reflected via measuring vibration [3,4], acoustic emission (AE) [5,6], temperature [7,8], and debris [9,10]. Among those
techniques, a widely used technique for fault detection is the vibration-based signature analysis, including time-domain
statistics [11,12], spectrum based analysis [13,14], and time-frequency representation [15,16].
Among the vibration signal based fault diagnosis algorithms, the kurtosis based method [17] is one of the most powerful
techniques for localized fault diagnosis. In specific, the resonance frequency band due to local defects related impulse series
is firstly located. And the band-pass filtering and envelope analysis methods are then used together to obtain the envelope
spectrum with prominent fault related frequency to achieve the final fault detection. During this strategy, the determination
of frequency band plays a key role, and several algorithms [18–22] has been proposed to accomplish this step. However,
most of the algorithms are proposed to detect mono-fault.
n
Corresponding author.
E-mail address: chufl@mail.tsinghua.edu.cn (F. Chu).
http://dx.doi.org/10.1016/j.jsv.2016.12.041
0022-460X/& 2017 Elsevier Ltd All rights reserved.
Please cite this article as: T. Wang, et al., Compound faults detection in gearbox via meshing resonance and spectral
kurtosis methods, Journal of Sound and Vibration (2017), http://dx.doi.org/10.1016/j.jsv.2016.12.041i
2 T. Wang et al. / Journal of Sound and Vibration ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Nomenclature bearing fault

xgearfault Gear fault related signal
xgearmeshing Gear meshing signal fmeshing Meshing frequency
xbearingfault Rolling bearing fault related signal fr,driven Rotational frequency of the driven gear
fr,driving Rotational frequency of the driving gear Bgear Magnitudes of the FM in gear meshing signal
Agear Magnitudes of the AM in gear meshing signal βmeshing Structural damping characteristic frequency of
Ameshing Amplitude of the impulses due to the meshing meshing behavior
behavior tn,meshing The occurrence time of the nth impulse due to
φmeshing Resonance frequency excited by the meshing the meshing behavior
behavior N Total number of impulses induced by the
n Serial number of the impulses induced by the meshing behavior
meshing behavior Agearfault Magnitudes of the AM in gear faulty signal
fr,gearfault Rotational frequency of the faulty gear Am,gearfault Amplitude of the impulses due to the gear
Bgearfault Magnitudes of the FM in gear faulty signal fault
βgearfault Structural damping characteristic frequency of ωgearfault Resonance frequency excited by the gear fault
gear fault m Serial number of the impulses induced by the
tm,gearfault The occurrence time of the mth impulse due to gear fault induced
the gear fault Abearingfault Amplitude of the impulses due to the bear-
M Total number of the impulses induced by the ing fault
gear fault ωbearingfault Resonance frequency excited by the bearing
βbearingfault Structural damping characteristic frequency fault
of bearing fault i Serial number of the impulses induced by the
τj random slippage bearing fault
I Total number of impulses induced by the n(t) White noise
Compound faults indicate that several faults appear simultaneously, or that one fault leads to the following fault.
Comparing with the traditional algorithms designed for the mono-fault condition, the methods for compound faults de-
tection are relatively fewer. Moreover, this compound faults are common in real application and may cause severe dis-
ruption. Towards this topic, McFadden et al. [23] proposed a model for the high-frequency vibration produced by multi-
point defects. Li et al. [24,25] employed principal component analysis and fuzzy k-nearest neighbor (FKNN) algorithms to
extract multi-fault feature in rotating machinery, respectively. Purushotham et al. [26] and Lei et al. [27] separately used
hidden Markov model (HMM) classifiers and hybrid intelligent method to determine multi-faults. Wang et al. [28] designed
an adaptive spectral kurtosis (SK) algorithm to identify multiple transient faults in the rolling bearings. Several wavelet
transform based algorithm [29–31] have been proposed for compound faults detection with good performance. Zhao et al.
[32] made use of a generalized demodulation algorithm to accomplish compound fault detection under time-varying ro-
tational speed. Some of the algorithms, [24–28,31,32] were proposed to detect faults that exist in a single component, such
as a rolling bearing or gearbox, and algorithms mentioned in [29,30] can deal with faults existing in different components
based on the approved multiwavelet transform. These contributions have significantly enriched the literature for the
compound faults diagnosis.
According to the algorithms summarized previously, a typical compound faults diagnosis strategy distinguishes the faults
and then locates the corresponding fault features, particular when multi-faults appear in different components. This study
proposes a new algorithm to detect compound faults that happen in the gear and the rolling bearing of a gearbox si-
multaneously. This algorithm can distinguish the resonance frequency bands induced by the gear and bearing faults by
taking full advantage of a commonly overlooked meshing resonance phenomenon [33,34], which indicates that the meshing
behaviors will generate resonance even without the present of a fault. The underlying principle of the proposed algorithm is
that the characteristics of the resonance frequency bands induced by the gear and rolling bearing faults are different from
each other. The proposed algorithm mainly consists of the following four steps: (a) decomposing the raw signal with qua-
sianalytic filters, (b) obtaining the kurtogram using the fault SK analysis, (c) calculating the MRgram based on the meshing
index, (d) detecting the compound faults by comparing the Kurtogram and the MRgram.
The rest of the paper are structured as follows: Section 2 describes the specific compound faults detection algorithm
based on meshing resonance and spectral kurtosis methods. The effectiveness of the proposed algorithm is separately
testified with simulated and experimental data in Sections 3 and 4. And the conclusions are drawn in Section 5.
2. Compound faults diagnosis algorithm based on meshing resonance and spectral kurtosis methods
As mentioned above, the key step of compound faults diagnosis is distinguishing the faults from each other. When
localized faults appear in both the gear tooth and the rolling bearing, the contacts between these two defects and their
T. Wang et al. / Journal of Sound and Vibration ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3
mating surfaces will result in two series of impulses. Moreover, these two impulse trains will both generate resonances at
relatively higher frequency bands simultaneously. However, it is difficult to distinguish the two impulse trains not only in
the time domain, but also in the resonance frequency bands.
To solve this problem, a meshing resonance (MR) phenomenon is considered. The MR phenomenon has three specific
characteristics which are: (1) the gear meshing behavior can result in the MR phenomenon, regardless of the presence of
faults. (2) the MR phenomenon will extend into different areas, if a localized fault is present. (3) the new gear fault induced
MR area is impulsive. These characteristics perhaps are significant in realizing the corresponding compound faults detection,
because these features are unique to the gearboxes. Considering that traditional kurtosis based algorithms can capture the
impulses generated by the localized faults of both the gear and rolling bearing, the compound faults diagnosis can be
realized if the MR frequency band induced by the gear fault can be identified.
In detail, the MR areas are captured by newly constructed meshing index and displayed in an MRgram. The resonance
frequency bands resulting from the localized faults can be determined using the spectral SK algorithm and shown in the Kur-
togram. In the MRgram, only the gear related resonance frequency bands can be located. In the kurtogram, both the resonance
frequency bands resulting from the localized faults of the gear and the rolling bearing can be identified. Hence, the compound
faults can be separately detected by comparing the MRgram and the Kurtogram via following rules: (1) The frequency band with
higher values (meshing index and SK value) in both the MRgram and the Kurtogram reflects the gear fault. (2) The frequency
band with higher values only in the Kurtogram, and not the MRgram indicates that the fault is in the rolling bearing.
It must be noted that some parts of the proposed algorithm related to the Kurtogram were proposed in a previous study[35].
The algorithm of the MRgram is original. The characteristics of the two algorithms are used together to diagnose compound faults.
In conclusion, a four-step algorithm is proposed to realize the compound faults diagnosis. Section 2.1 introduces the
signal composition algorithm. Sections 2.2 and 2.3 separately present the constructions the corresponding Kurtogram and
MRgram. The comparison based compound faults diagnosis is finally drawn. This proposed compound faults diagnosis
method is based on the meshing resonance and SK analyses, details of which are described in the following four sections.
2.1. Signal composing via quasianalytic filters
The vibration signal is initially decomposed into different filter bands and levels using a filtering algorithm described in
[21]. In general, this filtering algorithm originated from quasianalytic filters consisting of arborescent filter-bank structures.
First, quasianalytic low-pass and high-pass filters based on a low-pass prototype filter, h(n), with a cutoff frequency of fc
¼ ⅛ þ ε, ε Z 0 are constructed as follows:
h0( n) = h( n) × 2 jπn /4 (1)
h1( n) = h( n) × 2 j3πn /4 (2)
where h0(n) and h1(n) represent the low-pass and high-pass filters with normalized frequency bands of (0, ¼) and (¼, ½),
respectively.
The above process is repeated in a pyramidal manner to realize the binary tree decomposition. Ultimately, the signal is de-
composed into k levels, with each level having 2k bands. The iteration process between the adjacent levels is expressed as follows:
Ck2+i 1 = h0( m) × Cki( 2n) (3)
Ck2+i +11 = h1( m) × Cki( 2n) (4)

1 -k 1
where Ci k(n) is the complex envelope of the signal, x(n), based on the center frequency, fci ¼(iþ2 )2 , and the bandwidth,
Δfk ¼2-k 1. The parameter, |Ci k(n)|, is the corresponding envelope signal.
Finally, the ⅓-binary tree filter bank is added to obtain decomposition with higher resolution based on the existing levels.
The ⅓-binary decomposition is similar in principle to binary decomposition; its associated algorithm is detailed in a pre-
vious study [21].
2.2. Kurtogram obtainment [21]
In this section, the kurtosis values of the filtering results obtained in Section 2.1 are calculated to reveal the distribution
of the impulsive components of a signal in frequency domain. The SK of a signal x(t) can be defined as the normalized
fourth-order spectral moment [16], i.e.,
x4( t, f )
K x( f ) = 2
−2
X2( t , f ) (5)
where o4 represents the time-frequency averaging operator, and X(t, f) is the time-frequency envelope of the band-pass
filtered signal of x(t) around f. If the impulsive parts are concentrated more on a certain frequency band, the kurtosis of the
filtered signal will be accordingly higher. Hence, Kx(f) can be set as a criterion to find the most impulsive part of the signal in
the frequency domain.
With these kurtosis values, the kurtogram is constructed to display the kurtosis value calculated in Section 2.1. Based on
the kurtogram, the frequency band in which the transients take place can be located by identifying larger SK values. Fig. 7
(b) shows the appearance of the kurtogram.
If faults exit on the gear tooth and the rolling bearing simultaneously, relatively larger SK values can be found in cor-
responding resonance frequency bands.
2.3. MRgram calculation
As the MR phenomenon is a unique characteristic of gear meshing vibration, the determination of the MR area plays a
fundamental role in the success of the proposed method. In this section, the construction of a meshing index and an
MRgram is presented sequentially to determine and then display the MR areas with a reverse thought process.
Assuming that a frequency band belongs to the meshing resonance area, the meshing frequency should be modulated by
this frequency band. Consequently, the meshing frequency and its harmonic are prominent with less interruption in the
corresponding envelope spectrum. In other words, if the magnitudes of the gear meshing frequency and its harmonics are
much higher than the ones of other spectral lines in the envelope spectrum, the corresponding filter band can be considered
as the candidate of meshing resonance area.
To identify the MR area, a meshing index is constructed to describe the prominence of the gear meshing frequency and
its harmonics. This following index can be used to determine the MR area:
⎛ ⎞⎞
Index meshing = 1/Rank X envelope⎜ max(Xenvelope(f ) ⎟⎟ +
⎝ fmeshing − 0.5fbandwidth ≤ f ≤ fmeshing + 0.5fbandwidth ⎠⎠
⎛ ⎞⎞
⎝
(
1/Rank X envelope⎜ max Xenvelope (f ) ⎟⎟ +
2 × fmeshing − 0.5fbandwidth ≤ f ≤ 2 × fmeshing + 0.5fbandwidth ⎠⎠
⎛ ⎞⎞
1/Rank X envelope⎜⎜ max( X (f ) ⎟⎟
⎟⎟
envelope
⎝ 3 × fmeshing − 0.5fbandwidth ≤ f ≤ 3 × fmeshing + 0.5fbandwidth ⎠⎠ (6)
where Xenvelope represents the envelope spectrum of the filtering result x(t) obtained in Section 2.1. fmeshing is the meshing
frequency of the planetary gearbox. As the three parts of Eq. (6) are similar, only the first line is explained
⎛ ⎞
below.max⎜ Xenvelope(f ) ⎟ is the spectrum in the frequency range (fmeshing – 0.5fbandwidth,
⎝ fmeshing − 0.5fbandwidth ≤ f ≤ fmeshing + 0.5fbandwidth ⎠
⎛ ⎞
fmeshing þ 0.5fbandwidth), and max⎜ Xenvelope(f ) ⎟ is the highest amplitude. The para-
⎛ ⎛ ⎞⎞
meter,Rank X envelope⎜ max⎜ Xenvelope(f ) ⎟⎟ , represents the rank number of
⎝ ⎝ fmeshing − 0.5fbandwidth ≤ f ≤ fmeshing + 0.5fbandwidth ⎠⎠
⎛ ⎞
max⎜ Xenvelope(f ) ⎟ in Xenvelope( f) after all the data in Xenvelope ( f) are sorted in descending
order by magnitude. A higher value of Indexmeshing indicates a stronger gear related MR in the corresponding frequency
band. In this paper, the value of fbandwidth is set as 10 Hz.
To compare with the Kurtogram constructed in Section 2.2, the values of Indexmeshing calculated above in an MRgram are
displayed, whose appearance just resembles a Kurtogram but has a different practical application. A Kurtogram supports
the determination of the maximum impulsive frequency band, while the MRgram supports the selection of a gear's MR
frequency band. A frequency band with a higher meshing index can be considered to belong to an MR area.
2.4. Compound faults detection by comparing the Kurtogram and the MRgram
In this section, the compound faults detection is realized by comparing the Kurtogram and the MRgram in a three-step
process.
First, the rules to determine the resonance frequency areas (gear fault resonance (GFR) area and bearing fault resonance
(BFR) area) resulting from the faults located on different components based on the comparison between the Kurtogram and
the MRgram are outlined below.
(1) The frequency band having higher values of both meshing index and SK value in the MRgram and Kurtogram can be
seen as the mirage of the localized gear fault in the higher frequency area. In this study, the area concentrated with this
kind of frequency band is called the GFR area.
(2) The frequency band that contains only higher SK value in the Kurtogram can be considered as the bearing fault induced
band. The corresponding area with this type of frequency band is the BFR area.
Furthermore, the GFR frequency band and BFR frequency band are further selected in the GFR and BFR areas. In detail,
the frequency band with the highest meshing index in the GFR area of the MRgram is set as the GFR frequency band, and the
frequency band with the highest SK value in the BFR area of the Kurtogram is a proper BFR frequency band.
Ultimately, the compound faults are detected by filtering the raw signal with selected resonance frequency bands and
the envelope analysis.
In particular, the raw signal is filtered with the selected bands (the GFR and BFR frequency bands) in the first step, and
the Hilbert transform and fast Fourier transform (FFT) are used together to obtain the envelope spectra. The compound
faults can be finally detected via the envelope spectra.
Fig. 1 presents a flowchart of the proposed method for localized gear fault detection and diagnosis based on the MR
phenomenon. As illustrated in the flowchart, the Kurtogram and MRgram of the raw signal are separately calculated using
the algorithm mentioned in Section 2. The compound faults can be finally detected by comparing them.
Furthermore, the performance of the proposed algorithm is illustrated in Fig. 2 with a real vibrational signal of a gearbox
with two localized faults on both the gear tooth and the outer race of the rolling bearing.
3. Simulated analysis
In this section, the simulation is used to confirm the effectiveness of the proposed fault detection and diagnosis method.
In specific, the synthetic signal is contained with three parts: the gear meshing part (xgearmeshing), gear fault related part
(xgearfault) and the rolling bearing fault related part (xbearingfault).
As for the gear meshing part, it should reflect the MR phenomenon in a relatively higher frequency band. The corre-
sponding model is as follows:
xgearmeshing (t ) = [1 + Agear cos(2πfr, driving t )] cos [2πfmeshing t + Bgear sin(2πfr, driven t )] +
[1 + Agear cos(2πfr, driven t )] cos [2πfmeshing t + Bgear sin(2πfr, driven t )] +

N
∑ A meshing e−βme sin g (t − tn,meshing ) sin(φmeshing(t−tn, meshing ))u(t−tn, meshing )
n= 1 (7)
where the first two lines describe the amplitude modulation and frequency modulation (AMFM) phenomenon [36], and the
third line represents the impulse series caused by the gear meshing behavior. Specifically, fmeshing is the meshing (carrier)
frequency, fr,driving and fr,driven are the rotational (modulated) frequencies of the driving and driven gears, and Agear and Bgear
are the magnitudes of the AM and FM, respectively. In the third line, Ameshing is the amplitude of the impulses caused by the
meshing behavior, βmeshing is the structural damping characteristic frequency, φmeshing is the excited resonance frequency, tn,
th
meshing is the occurrence time of the n impulse.
As for the part representing the localized gear fault, the model is expressed in Eq. (8).
xgearfaults (t ) = [1 + Agearfault cos(2πfr, fault t )] cos [2πfmeshing t + Bgearfault sin(2πfr, fault )] +

M
∑ A m, gearfault e−βgearfault (t − tm, gearfault ) sin(ωgearfault (t−tm, gearfault ))u(t−tm, gearfault )+
m=1 (8)
where, the first line describes the new AMFM phenomenon because of the gear fault, and the second line gives the impulse
series resulting from a localized gear fault. In particular, fmeshing is the meshing (carrier) frequency of the gearbox and fr,fault is
the rotational (modulated) frequency of the faulty gear. Agearfault, and Bgearfault are the magnitudes of the AM and FM in the
gear fault signal, respectively. The parameters Am,gearfault, βgearfault, ωgearfault and tm,gearfault, are the counterparts of An,meshing,
βmeshing, ωmeshing and tn,meshing for the gear meshing part in Eq. (7).
Eq. (9) indicates the impulses generated by the bearing fault [37].
Fig. 1. Flowchart of the proposed method for compound faults diagnosis.
Fig. 2. The framework of the proposed algorithm for detection of compound faults using a real signal with a gear pitting fault and a rolling bearing outer
race fault.
⎛ i ⎞
I ⎜
−βbearingfault ⎜ t − ti, bearingfault − ∑ τj⎟
⎟ ⎛ ⎛ i ⎞⎞
⎜ ⎟
xbearingfault (t ) = ∑ Abearingfault e ⎝ j =1 ⎠ sin⎜ ωbearingfault ⎜⎜ t−ti, bearingfault − ∑ τj⎟⎟⎟⎟
⎜
i=1 ⎝ ⎝ j=1 ⎠⎠
⎛ i ⎞
u ⎜⎜ t−ti, bearingfault − ∑ τj⎟⎟
⎝ j=1 ⎠ (9)
where Abearingfault, βbearingfault, ωbearingfault, and tn, bearingfault are the counterparts of An,meshing, βmeshing, ωmeshing, and tn,meshing. τi
represents the effect of random slippage of the rolling elements. Its value is set as the product of 0.01 0.02 and the time
interval between two rolling bearing fault induced impulses.
In the three equations aforementioned, the parameters tn,meshing, tm,gearfault, and ti, bearingfault can be calculated using the
following equations.
tn, meshing = n/fmeshing (10)
tm, gearfault = m/fr, gearfault (11)
ti, bearingfault = i/fbearingfault (12)
where fr, gearfault is the rotational frequency of the shaft with faulty gear. f bearingfault indicates the fault characteristic fre-
quency of the faulty rolling bearing.
Hence, the synthetic signal of the gearbox with compound faults can be obtained using the following equation.
xcompoundfaults (t ) = x meshinggear (t ) + xgearfault (t ) + xbearingfault (t ) + n(t ) (13)
where n(t) is the white noise. The definition of all the parameters used above are listed in Table 1.
In this study, the rotational speed of the input gear was set as 1000 rpm and, the other parameters of the model are listed
in Table 2. In here, the gear-fault induced resonance frequency is set higher than the one induced by the gear meshing. It can
be partly explained using Fig. 8(e), which is the MRgram of a measured signal with compound faults (gear tooth pitting fault
and bearing inner race fault) given in Section 4.2. In the MRgram, the area with higher meshing indexes near 5000 Hz is
influenced by the gear meshing, and the one at approximately 10000 Hz is induced by the gear fault. To further explain this
setting, a comparison experiment is included in the Appendix A.
Fig. 3(a) shows the simulated vibration signal in the time domain. The proposed algorithm is applied to analyze the
simulated signal. The results are plotted in Fig. 3(b) – (g).
First, the GFR and BFR areas can be determined by comparing the corresponding kurtogram and MRgram shown in Fig. 3
(b) and (c) by following the three observations given below.
(1) It can be seen that the area around 5000 Hz not only has larger SK values in the Kurtogram (in Fig. 3(b)) but also has
larger Meshing indexes in the MRgram (in Fig. 3(c)). In this way, this area is set as the GFR area which is marked with
solid black rectangles.
(2) Another area with relatively higher SK values around 7000 Hz can be identified in the kurtogram (in Fig. 3(c)), and the
corresponding area in the MRgram has no higher meshing indexes. This area can be considered as the BFR area which is
marked with a dashed black rectangle.
(3) In the MRgram, the area with higher meshing indexes around 2000 Hz is due to the gear meshing behavior.
Table 1
Parameter definitions of the simulation model.
xgearmeshing Gear meshing signal xgearfault Gear fault related signal
xbearingfault Rolling bearing fault related signal fmeshing Meshing frequency

fr,driving Rotational frequency of the driving gear fr,driven Rotational frequency of the driven gear
Agear Magnitudes of the AM in gear meshing signal Bgear Magnitudes of the FM in gear meshing signal
Ameshing Amplitude of the impulses due to the meshing behavior βmeshing Structural damping characteristic frequency of the meshing
behavior
φmeshing Resonance frequency excited by the meshing behavior tn,meshing The occurrence time of the nth impulse due to the meshing
behavior
n Serial number of the impulses induced by the meshing N Total number of the impulses induced by the meshing behavior
behavior
fr,gearfault Rotational frequency of the faulty gear Agearfault Magnitudes of the AM in gear faulty signal
Bgearfault Magnitudes of the FM in gear faulty signal Am,gearfault Amplitude of the impulses due to the gear fault
βgearfault Structural damping characteristic frequency of gear fault ωgearfault Resonance frequency excited by the gear fault
tm,gearfault The occurrence time of the mth impulse due to the gear m Serial number of the impulses induced by the gear fault
fault
M Total number of the impulse induced by the gear fault Abearingfault Amplitude of the impulses due to the bearing fault
βbearingfault Structural damping characteristic frequency of bearing ωbearingfault Resonance frequency excited by the bearing fault
fault
τj random slippage i Serial number of the impulses induced by the bearing fault
I Total number of the mpulses induced by the bearing fault n(t) White noise
Table 2
Simulation model parameters.
Agear Bgear fmeshing fr,driving fr,driven Ameshing
0.2 0.3 367 Hz 16.67 Hz 6.67 Hz 0.5

βmeshing φmeshing Agearfault Bgearfault fr,gearfault Am,gearfault
1200 2000 Hz 0.1 0.3 16.67 Hz 2
βgearfault ωgearfault βgearfault ωbearingfault Abearingfault βbearingfault
1200 5500 Hz 1200 7000 Hz 0.4 1200
Fig. 3. Simulated compound faults: (a) raw signal, (b) Kurtogram, (c) MRgram, (d) Kurtogram with the GFR and BFR frequency bands, (e) MRgram with the
GFR frequency band, (f), Envelope spectrum based on the GFR frequency band, (g) Envelope spectrum based on the BFR frequency band.
Fig. 4. Details of the synthetic signal of the compound faults: (a) gear meshing signal, (b) gear fault signal, (c) bearing fault signal.
In Table 1, the resonance frequencies of the gear and bearing faults are preset to values 5000 Hz and 7000 Hz, respec-
tively, both of which located in the selected GFR and BFR areas using the proposed algorithm. This can partly testify the
corresponding effectiveness.
Thereafter, the GFR and the BFR frequency bands are selected from the GFR and BFR areas by adhering to the rules
mentioned in Section 2.4. In specific, the GFR frequency band is [4778.7 Hz, 5461.3 Hz] in level 3.5. The BFR frequency band
is [6656 Hz, 7168 Hz] in level 4.
At last, the compound faults can be both detected by filtering the raw signal with the GFR and BFR frequency bands.
Fig. 3(f) and (g) show the corresponding envelope spectrums of the filtered results of the GFR and the BFR frequency bands,
respectively. In Fig. 3(f), the rotational frequency of the faulty gear can be located both in the lower frequency area and
around the meshing frequency. In Fig. 3(g), prominent peaks can be easily found at 59.5 Hz, 119 Hz, 178.5 Hz, 238 Hz and
297.5 Hz. They almost equal the preset bearing fault characteristic frequency and its harmonics.
In conclusion, the two envelope spectrums can finally indicate the presence of the localized gear and bearing faults. This
proves that the proposed algorithm is effective.
To provide the details of the simulated signal used above, the main parts, xmeshingear, xgeaffault and xbearingfault, are separately
given in Fig. 4.
4. Experimental tests
In this section, the performance of the proposed compound faults diagnosis algorithm is further testified using ex-
perimental trials. Two types of compound faults cases are separately carried out based on a gearbox test rig at Tsinghua
University. One instance is that localized faults appear in both the gear tooth and the outer race of the rolling bearing, and
the other one is that localized faults appear in both gear tooth and the inner race of the rolling bearing. The experimental
setup is shown in Fig. 5. A motor which is controlled by an alternating current (AC) inverter was used to drive a single-stage
gearbox. A tachometer and accelerometers were respectively mounted on the test rig to measure the motor rotational speed
and the vibration signal. A magnetic powder brake was connected to the output shaft to induce a workload. The vibration
signals measured by the sensor placed on the left side are used in this section.
The localized faults on the gear and the rolling bearing are manually included on the tooth surface and outer and inner
races of the bearing, respectively. For the gear pitting fault, the corresponding shape is a circle with a radius of 1 mm. And
the bearing faults are added by cutting slide on the bearing outer race and the inner race with a width of 0.5 mm. Fig. 6(a–c)
Fig. 5. Experimental set-up for gear fault detection and diagnosis.
Fig. 6. Localized faults: (a) gear pitting fault, (b) rolling bearing outer race fault, and (c) rolling bearing inner race fault.
Table 3
Test gearbox parameters.
Number Number Tooth Tooth Gear ratio Module

of teeth of teeth width in width in
in driving in driven driving driven
gear gear gear gear
22 55 27 mm 24 mm 2.5 1.5
Bearing Number Pitch Ball fcfouter fcfinner
type of balls diameter diameter
6206 9 46 mm 10.4 mm 3.57*fr 5.43*fr
illustrate gear-tooth pitting fault and faults in the outer and inner races of the rolling bearing, respectively.
The gear with the pitting fault is installed on the input shaft, and the faulty bearings are mounted on the output shaft.
Table 3 lists the parameters of the gearbox.
4.1. Case 1: Gear tooth with pitting fault and rolling bearing with outer race fault
In this section, the proposed algorithm is validated using experimental data of a gearbox with compound faults. There
exists a pitting fault on a tooth of the driven gear (seen in Fig. 6(a)), and outer race fault on a rolling bearing supporting the
input shaft (seen in Fig. 6(b)). The corresponding rotational speed is 1000 rpm. According to the parameters listed in Table 3,
the meshing frequency is 367.1 Hz; the rotational frequencies of the input and output shafts are 16.67 Hz and 6.67 Hz,
respectively. As for the fault characteristic frequencies, the rotational frequency of the faulty gear is 6.67 Hz, and the fault
characteristic frequency (fcf) of the faulty rolling bearing is 59.51 Hz. The sampling rate is 32768 Hz, the data is collected for
15 s, and the raw vibration signal is presented in Fig. 7(a).
With the algorithm proposed above, the Kurtogram and MRgram are separately calculated and illustrated in Fig. 7(b) and
(c), respectively. In the Kurtogram, two areas (one locates around 13000 Hz, and the other locates around 10000 Hz.) with
relatively higher SK values can be located. And as for the MRgram, two areas with higher meshing indexes are identified.
One of them is around 3500 Hz, and another can be found around 10000 Hz. By comparing these two figures, three
Fig. 7. Gear tooth with pitting fault and rolling bearing with outer race fault: (a) raw signal, (b) Kurtogram, (c) MRgram, (d) Kurtogram with the GFR
frequency band and the BFR frequency band, (e) MRgram with the GFR frequency band, (f), Envelope spectrum based on the GFR frequency band,
(g) Envelope spectrum based on the BFR frequency band.
Fig. 8. Gear tooth with pitting fault and rolling bearing with inner race fault: (a) raw signal, (b) Kurtogram, (c) MRgram, (d) Kurtogram with the GFR
frequency band and the BFR frequency band, (e) MRgram with the GFR frequency band, (f), Envelope spectrum based on the GFR frequency band,
(g) Envelope spectrum based on the BFR frequency band.
conclusions can be drawn as follows:
(1) The areas around 10000 Hz can be both found in corresponding Kurtogram and MRgram with relatively higher SK
values and meshing indexes. More important, the shapes of them are quite similar, which makes this area the GFR area.
Two solid polygons are used to mark this band, as shown in Fig. 7(d) and (e).
(2) In the Kurtogram, another prominent area with higher SK values is the BFR area. A dashed rectangle is used to mark this
BFR band in Fig. 7(d). It is worth mentioning that this area cannot be found in the MRgram shown in Fig. 7(e).
(3) In the MRgram, the area around 3500 Hz is due to the gear meshing behaviors. Moreover, this area is not observed for
higher SK values in the corresponding Kurtogram.
Based on the GFR and BFR areas obtained above, the GFR and BFR frequency bands are separately selected according to the
rules constructed in Section 2.4. In specific, the GFR frequency band is [9216 Hz, 10240 Hz] in level 4. The BFR frequency band is
[12288 Hz, 13653.3 Hz] in level 3.5. To finally detect the compound faults, the raw signal is separately band-filtered with the
selected resonance frequency bands, and the corresponding envelope spectrums are shown in Fig. 7(f) and (g). As for the
envelope spectrum shown in Fig. 7(f), prominent peaks related to the rotational frequency of the faulty gear, fr,fault, are clearly
identified. In the lower frequency zone, peaks representing fr,fault, 2 fr,fault, and 3 fr,fault are relatively prominent. Considering
the sidebands around the meshing frequency, prominent peak spacing at fr,fault and 2 fr,fault were also identified. These
characteristics can verify the correctness of the selected GFR frequency band. In Fig. 7(g), prominent peaks at the fcfouter and its
harmonics are evident. Therefore, it can be diagnosed that the fault is in the outer race of the bearing supporting the input shaft.
In conclusion, the proposed algorithm is able to separate the compound localized faults in the gear tooth and the rolling
bearing and then realized the corresponding fault diagnosis.
4.2. Case 2: gear tooth with pitting fault and rolling bearing with inner race fault
In this case, most of the operational parameters are as the same as the ones in case 1. The only difference is that the
rolling bearing fault locates in the corresponding inner race. Fig. 8(a) (f) are the counterparts of the Fig. 7(a) (f).
With the proposed algorithm, the Kurtogram and the MRgram of the raw signal (shown in Fig. 8(a)) are calculated and
shown in Fig. 8(b) and (c), respectively. By comparing these two figures, the BFR and GFR areas and then the BFR and GFR
frequency bands are separately obtained. In detail, the area marked by a solid black rectangle in the Kurtogram shown in
Fig. 8(d) is the BFR area. The areas marked by dashed black rectangles in both the Kurtogram and the MRgram in Fig. 8
(d) and (e) represent the GFR area. The corresponding BFR and GFR frequency bands are [13312 Hz, 14336 Hz] in level 4 and
[9557.4 Hz, 10922.6 Hz] in level 3.5. Based on these select frequency bands, the raw signal is finally filtered, and the envelope
spectrums are shown in Fig. 8(f) and (g).
In Fig. 8(f), the spectral lines related to the gear localized fault can be found in the lower frequency area and around the
meshing frequency. In Fig. 8(g), prominent spectral peaks representing the inner race fault characteristic frequency (90.5 Hz)
and its harmonics can be easily located.
According to the above analysis, the performance of the proposed algorithm can be further testified using the compound
faults of the gear tooth pitting fault and the inner race fault of the rolling bearing.
5. Conclusion
A compound faults diagnosis algorithm has been proposed based on the meshing resonance and SK methods. This
algorithm can divide the localized gear and rolling bearing faults by comparing the Kurtogram and the MRgram.
The major innovative features of the proposed method are reflected by: (a) the gear meshing resonance area identifi-
cation using the meshing index and MRgram, (b) the BFR and GFR frequency bands determination based on the comparison
between the MRgram and Kurtogram, and (c) compound faults diagnosis via envelope analysis. The simulation and ex-
perimental tests have demonstrated that the proposed method is effective in diagnosing the localized gear and rolling
bearing faults simultaneously.
Acknowledgement
The research work described in this paper was supported by Natural Science Foundation of China under Grant no.
51335006 and 51605244. The valuable comments and suggestions from the reviewers are very much appreciated.
Appendix A. A comparison experimental case
In this section, a comparison experimental case is considered to explain why the resonance frequencies induced by the
gear meshing and the gear fault are set to different values, as seen in Section 3.
Fig. A1. Comparison experimental case: (a) waveform of the healthy vibration signal, (b) waveform of the gear fault vibration signal, (c) MRgram of the
healthy vibration signal, (d) MRgram of the gear fault vibration signal.
In specific, two signals, the health condition and fault condition are measured by the accelerate mounted on the test rig
shown in Fig. 5. In this section, only one fault exists on the gear tooth. Fig. A1(a) and (c) represent the raw waveform and the
MRgram of the healthy vibration signal, respectively. Fig. A1(b) and (d) separately show the waveform and the MRgram of
the gear fault vibration signal, respectively. By comparing these two MRgrams, two conclusions can be drawn.
First, the frequency area with higher meshing indexes around 3500 Hz appears, as seem in both Fig. A1(c) and (d), which
indicates that this frequency area is generated by the gear meshing. Then, the frequency area with higher meshing indexes
around 12000 Hz appears only in Fig. A1(d), implying that this frequency area is generated by the gear fault.
According to the above analysis, the resonance frequency induced by the gear fault is set higher than the one caused by
gear meshing, seen in Section 3.
References
[1] R.B. Randall, J. Antoni, Rolling element bearing diagnostics-a tutorial, Mech. Syst. signal Process. 25 (2) (2011) 485–520.
[2] P.D. Samuel, D.J. Pines, A review of vibration-based techniques for helicopter transmission diagnostics, J. Sound Vib. 282 (1–2) (2005) 475–508.
[3] T. Wang, et al., Rolling element bearing fault diagnosis via fault characteristic order (FCO) analysis, Mech. Syst. Signal Process. 45 (1) (2014) 139–153.
[4] G. Dalpiaz, A. Rivola, R. Rubini, Effectiveness and sensitivity of vibration processing techniques for local fault detection in gears, Mech. Syst. Signal
Process. 14 (3) (2000) 387–412.
[5] C.K. Tan, D. Mba, Identification of the acoustic emission source during a comparative study on diagnosis of a spur gearbox, Tribol. Int. 38 (5) (2005)
469–480.
[6] N. Tandon, A. Choudhury, A review of vibration and acoustic measurement methods for the detection of defects in rolling element bearings, Tribol. Int.
32 (8) (1999) 469–480.
[7] A.M.D. Younus, B. Yang, Intelligent fault diagnosis of rotating machinery using infrared thermal image, Expert Syst. Appl. 39 (2) (2012) 2082–2091.
[8] V.T. Tran, et al., Thermal image enhancement using bi-dimensional empirical mode decomposition in combination with relevance vector machine for
rotating machinery fault diagnosis, Mech. Syst. Signal Process. 38 (2) (2013) 601–614.
[9] Z. Peng, N. Kessissoglou, An integrated approach to fault diagnosis of machinery using wear debris and vibration analysis, Wear 255 (7–12) (2003)
1221–1232.
[10] T.H. Loutas, et al., The combined use of vibration, acoustic emission and oil debris on-line monitoring towards a more effective condition monitoring of
rotating machinery, Mech. Syst. Signal Process. 25 (4) (2011) 1339–1352.
[11] Y. Lei, et al., Two new features for condition monitoring and fault diagnosis of planetary gearboxes, J. Vib. Control 0 (0) (2013) 1–10.
[12] F. Li, et al., Wavelet transform-based higher-order statistics for fault diagnosis in rolling element bearings, J. Vib. Control 14 (11) (2008) 1691–1709.
[13] T. Wang, et al., Bearing fault diagnosis under unknown variable speed via gear noise cancellation and rotational order sideband identification, Mech.
Syst. Signal Process. 62–63 (2015) 30–53.
[14] W. Wang, Early detection of gear tooth cracking using the resonance demodulation technique, Mech. Syst. Signal Process. 15 (5) (2001) 887–903.
[15] C. Li, M. Liang, Time–frequency signal analysis for gearbox fault diagnosis using a generalized synchrosqueezing transform, Mech. Syst. Signal Process.
26 (2012) 205–217.
[16] Z.K. Peng, P.W. Tse, F.L. Chu, A comparison study of improved Hilbert–Huang transform and wavelet transform: application to fault diagnosis for rolling
bearing, Mech. Syst. Signal Process. 19 (5) (2005) 974–988.
[17] Y. Wang, et al., Spectral kurtosis for fault detection, diagnosis and prognostics of rotating machines: a review with applications, Mech. Syst. Signal
Process. 66–67 (2016) 679–698.
[18] L.A.D.Y. Jiesi, A kurtosis-guided adaptive demodulation technique for bearing fault detection based on tunable-Q wavelet transform, Meas. Sci. Technol.
24 (5) (2013) 055009.
[19] Y. Wang, M. Liang, An adaptive SK technique and its application for fault detection of rolling element bearings, Mech. Syst. Signal Process. 25 (5) (2011)
1750–1764.
[20] Y. Lei, et al., Application of an improved kurtogram method for fault diagnosis of rolling element bearings, Mech. Syst. Signal Process. 25 (5) (2011)
1738–1749.
[21] J. Antoni, Fast computation of the kurtogram for the detection of transient faults, Mech. Syst. signal Process. 21 (1) (2007) 108–124.
[22] T. Wang, et al., A new SKRgram based demodulation technique for planet bearing fault detection, J. Sound Vib. (2016).
[23] P.D. McFadden, J.D. Smith, The vibration produced by multiple point defects in a rolling element bearing, J. Sound Vib. 98 (2) (1985) 263–273.
[24] Z. Li, et al., Virtual prototype and experimental research on gear multi-fault diagnosis using wavelet-autoregressive model and principal component
analysis method, Mech. Syst. Signal Process. 25 (7) (2011) 2589–2607.
[25] Z. Li, et al., Blind vibration component separation and nonlinear feature extraction applied to the nonstationary vibration signals for the gearbox
multi-fault diagnosis, Measurement 46 (1) (2013) 259–271.
[26] V. Purushotham, S. Narayanan, S.A.N. Prasad, Multi-fault diagnosis of rolling bearing elements using wavelet analysis and hidden Markov model based
fault recognition, NDT E Int. 38 (8) (2005) 654–664.
[27] Y. Lei, Z. He, Y. Zi, Application of a novel hybrid intelligent method to compound fault diagnosis of locomotive roller bearings, J. Vib. Acoust. 130 (3)
(2008) 034501.
[28] Y. Wang, M. Liang, Identification of multiple transient faults based on the adaptive spectral kurtosis method, J. Sound Vib. 331 (2) (2012) 470–486.
[29] J. Chen, et al., Compound faults detection of rotating machinery using improved adaptive redundant lifting multiwavelet, Mech. Syst. Signal Process. 38
(1) (2013) 36–54.
[30] S. He, et al., Multifractal entropy based adaptive multiwavelet construction and its application for mechanical compound-fault diagnosis, Mech. Syst.
Signal Process. 76–77 (2016) 742–758.
[31] Y. Wang, Z. He, Y. Zi, Enhancement of signal denoising and multiple fault signatures detecting in rotating machinery using dual-tree complex wavelet
transform, Mech. Syst. Signal Process. 24 (1) (2010) 119–137.
[32] D. Zhao, et al., Compound faults detection of rolling element bearing based on the generalized demodulation algorithm under time-varying rotational
speed, J. Sound Vib. 378 (2016) 109–123.
[33] V. Atanasiu, I. Doroftei, Dynamic contact loads of spur gear pairs with addendum modifications, Eur. J. Mech. Environ. Eng. 2 (2008) 21.
[34] K. Umezawa, T. Suzuki, T. Sato, Vibration of power transmission helical gears: approximate equation of tooth stiffness, Bull. JSME 29 (251) (1986)
1605–1611.
[35] J. Antoni, The spectral kurtosis: a useful tool for characterising non-stationary signals, Mech. Syst. signal Process. 20 (2) (2006) 282–307.
[36] Z. Feng, M.J. Zuo, Vibration signal models for fault diagnosis of planetary gearboxes, J. Sound Vib. 331 (22) (2012) 4919–4939.
[37] M. Liang, I. Soltani Bozchalooi, An energy operator approach to joint application of amplitude and frequency-demodulations for bearing fault de-
tection, Mech. Syst. signal Process. 24 (5) (2010) 1473–1494.

Compound Faults Detection in Gearbox Via Meshing Resonance and Spectral Kurtosis Methods.2016.12.041

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Compound Faults Detection in Gearbox Via Meshing Resonance and Spectral Kurtosis Methods.2016.12.041

Uploaded by

Copyright:

Available Formats

Journal of Sound and Vibration ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Journal of Sound and Vibration

Compound faults detection in gearbox via meshing resonance

Nomenclature bearing fault

2.1. Signal composing via quasianalytic filters

h0( n) = h( n) × 2 jπn /4 (1)

h1( n) = h( n) × 2 j3πn /4 (2)

Ck2+i 1 = h0( m) × Cki( 2n) (3)

Ck2+i +11 = h1( m) × Cki( 2n) (4)

2.2. Kurtogram obtainment [21]

2.3. MRgram calculation

xgearmeshing (t ) = [1 + Agear cos(2πfr, driving t )] cos [2πfmeshing t + Bgear sin(2πfr, driven t )] +

[1 + Agear cos(2πfr, driven t )] cos [2πfmeshing t + Bgear sin(2πfr, driven t )] +

xgearfaults (t ) = [1 + Agearfault cos(2πfr, fault t )] cos [2πfmeshing t + Bgearfault sin(2πfr, fault )] +

Fig. 1. Flowchart of the proposed method for compound faults diagnosis.

tn, meshing = n/fmeshing (10)

tm, gearfault = m/fr, gearfault (11)

ti, bearingfault = i/fbearingfault (12)

xgearmeshing Gear meshing signal xgearfault Gear fault related signal

xbearingfault Rolling bearing fault related signal fmeshing Meshing frequency

Agear Bgear fmeshing fr,driving fr,driven Ameshing

0.2 0.3 367 Hz 16.67 Hz 6.67 Hz 0.5

Fig. 5. Experimental set-up for gear fault detection and diagnosis.

Number Number Tooth Tooth Gear ratio Module

conclusions can be drawn as follows:

Appendix A. A comparison experimental case

You might also like