Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

Contents lists available at SciVerse ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Planetary gearbox fault diagnosis using an adaptive stochastic

resonance method
Yaguo Lei a,b,n, Dong Han a, Jing Lin a, Zhengjia He a
a
State Key Laboratory for Manufacturing Systems Engineering, Xi’an Jiaotong University, No. 28 Xianning West Road, Xi’an 710049, China
b
State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing 400044, China

a r t i c l e i n f o abstract

Article history: Planetary gearboxes are widely used in aerospace, automotive and heavy industry
Received 4 November 2011 applications due to their large transmission ratio, strong load-bearing capacity and high
Received in revised form transmission efficiency. The tough operation conditions of heavy duty and intensive
12 June 2012
impact load may cause gear tooth damage such as fatigue crack and teeth missed etc.
Accepted 22 June 2012
The challenging issues in fault diagnosis of planetary gearboxes include selection of
sensitive measurement locations, investigation of vibration transmission paths and
Keywords: weak feature extraction. One of them is how to effectively discover the weak
Planetary gearboxes characteristics from noisy signals of faulty components in planetary gearboxes. To
Adaptive stochastic resonance
address the issue in fault diagnosis of planetary gearboxes, an adaptive stochastic
Weak feature extraction
resonance (ASR) method is proposed in this paper. The ASR method utilizes the
Fault diagnosis
optimization ability of ant colony algorithms and adaptively realizes the optimal
stochastic resonance system matching input signals. Using the ASR method, the noise
may be weakened and weak characteristics highlighted, and therefore the faults can be
diagnosed accurately. A planetary gearbox test rig is established and experiments with
sun gear faults including a chipped tooth and a missing tooth are conducted. And the
vibration signals are collected under the loaded condition and various motor speeds.
The proposed method is used to process the collected signals and the results of feature
extraction and fault diagnosis demonstrate its effectiveness.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Planetary gearboxes are commonly used in aerospace, automotive and heavy industry applications to derive benefits
from their large transmission ratio, strong load-bearing capacity and high transmission efficiency [1]. Despite these
advantages, the tough operation conditions of heavy duty and intensive impact load may cause gear tooth damage such as
fatigue crack and teeth missed etc., occurrence in the planetary gearboxes frequently [2]. The tooth damage may lead to
undesirable dynamic behaviors resulting in large vibration, heavy noise and unacceptable performance of planetary
gearboxes.
The term planetary gearbox is actually a gearbox including a planetary gear set, which refers to the compound gear
systems with several planet gears between a center sun gear and an outer ring gear. Fig. 1 shows an elementary planetary

n
Corresponding author at: State Key Laboratory for Manufacturing Systems Engineering, Xi’an Jiaotong University, No. 28 Xianning West Road,
Xi’an 710049, China
E-mail address: yaguolei@mail.xjtu.edu.cn (Y. Lei).

0888-3270/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ymssp.2012.06.021

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
2 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

Ring gear

Planet gears

Sun gear

Fig. 1. A schematic showing the components of a 4-planet planetary gear set.

gear set with four planets. It is evident that an elementary planetary gear set has a stationary (non-rotating) ring gear, a
sun gear that rotates around its own center, and several planet gears that not only rotate around their own centers but also
revolve around the center of the sun gear. The planet gears mesh simultaneously with the sun gear and the ring gear.
A planetary gearbox is more complicated than a fixed-axis gearbox in gear transmission structures. In a planetary
gearbox, there are several planet gears that not only rotate around their own centers but also revolve around the center of
the sun gear. However, in a fixed-axis gearbox, all gears rotate around their own fixed centers only. Thus, a planetary
gearbox generally exhibits different behaviors from a fixed-axis gearbox. (1) In a planetary gearbox transmission system,
there are multiple and time-varying vibration transmission paths from gear meshing points to transducers, which are
typically mounted on the housing of a planetary gearbox. These vibration transmission paths may deteriorate or attenuate
vibration response of faulty gears through dissipation and interference effects [3]. In addition, torques or loads applied to
the gearbox may also lead to non-linear transmission path effects [4]. These effects would weaken the fault characteristics
hidden in vibration signals. (2) Planetary and fixed-axis gearboxes have different distributions in the frequency spectrum
of vibration signals. For a pair of meshing gears in a fixed-axis gearbox with gear faults, the fault characteristic frequencies,
sidebands, symmetrically locate around the meshing frequency and its harmonics in the frequency spectrum. For a
planetary gearbox, the sidebands are not symmetric about gear meshing frequency and its harmonics because multiple
planet gears produce similar vibrations but with different meshing phases, which cause some of the excitations of multiple
gear meshes to be canceled or neutralized [5–7]. (3) Some components in a planetary gearbox generally operate under low
rotating speeds because of the large transmission ratio. It is true that low-frequency characteristics are easily buried by
heavy noise. Thus, it is particularly difficult to discover the fault characteristics of low-speed components in the planetary
gearbox. The above three behaviors integrate to largely increase the difficulty of feature extraction and fault diagnosis of
the planetary gearbox.
A few interesting studies regarding condition monitoring and fault diagnosis of planetary gearboxes have been reported
recently in the literature. Blunt and Keller [4] developed two methods based on a planet and the carrier for detecting a crack
in the planetary carrier of a helicopter planetary gearbox. Bartelmus and Zimroz [8,9] investigated the influence of the
varying load on condition monitoring of planetary gearboxes and proposed a feature for monitoring planetary gearboxes
under time-variable operating conditions. Lei et al. [10,11] introduced methods based on new diagnostic parameters and
multi-sensor information fusion to classify different damage modes in a planetary gearbox. Barszcz and Randall [12] applied
the spectral kurtosis technique to crack detection of the ring gear in the planetary gearbox of a wind turbine. Zhang et al.
[13,14] integrated a blind deconvolution algorithm and vibration modeling for detecting a seeded fault in the helicopter
planetary gearbox. Samuel and Pines [15,16] proposed a technique based on the constrained adaptive lifting algorithm for
detecting gear fault with all teeth spalled in the planetary gearbox of a helicopter transmission. Zimroz and Bartelmus [17]
investigated the cyclo-stationary properties of vibration signals for multi fault detection in multi-stage gearboxes: fixed axis
and planetary. Bartkowiak and Zimroz [18] found outliers in a large data of two excavator planetary gearboxes under good
and bad condition respectively, and used them for condition monitoring of planetary gearboxes. Bartkowiak and Zimroz [19]
utilized data projection techniques to reduce observation space of real vibration signals from a mine excavator planetary
gearbox and explored the data distribution and dimensionality. In addition, Inalpolat and Kahraman, and Mark [20–23]
predicted the vibration and spectra of planetary gear transmission by establishing dynamic models. The above studies have
provided critical insight on fault diagnosis of planetary gearboxes. Investigations on fault diagnosis of planetary gearboxes,
however, are much limited in comparison with those of fixed-axis ones. There are still lots of issues to be addressed in fault
diagnosis of planetary gearboxes. The issues in fault diagnosis of planetary gearboxes include selection of sensitive
measurement locations, investigation of vibration transmission paths and weak feature extraction. One of them is how to
effectively discover the weak characteristics from noisy signals of faulty components in planetary gearboxes. Weak
characteristics mean small amplitudes of the spectral components in this paper.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]] 3

Stochastic resonance (SR) [24] has been proved as an effective tool in processing signals with low signal-to-noise ratio
by utilizing noise to enhance signal weak characteristics. However, the performance of SR definitely depends on system
parameters a and b. This paper proposes an adaptive stochastic resonance (ASR) method. The method incorporates the
merits of SR utilizing noise to enhance signal weak characteristics [24–26]. On the other hand, by utilizing ant colony
algorithms to optimize SR parameters, it overcomes the disadvantage of SR subjectively selecting parameters. The ASR
method is therefore expected to perform well in extracting weak characteristics for fault diagnosis of planetary gearboxes.
A two-stage planetary gearbox test rig is established and experiments on the test rig are carried out to verify the
effectiveness of the proposed method. Gears with a chipped tooth and a missing tooth are respectively installed in the test
rig and vibration signals are collected under the loaded condition with the value of 13.5 nm and various motor speeds. The
diagnosis results validate that the ASR method is able to discover the weak characteristics and diagnose the gear faults
effectively.

2. Experimental system and data acquisition

In order to demonstrate the effectiveness of the proposed ASR method in fault diagnosis of planetary gearboxes, a
planetary gearbox test rig is established and experiments on it are conducted. The experimental system is shown in
Fig. 2(a) and the simulated model of the gearbox test rig in Fig. 2(b). The test rig includes two gearboxes, a 3-hp motor for
driving the gearboxes, and a magnetic brake for loading. The motor rotating speed is controlled by a speed controller,
which allows the tested gear to operate under various speeds. The load is provided by the magnetic brake connected to the
output shaft and the load can be adjusted by a brake controller. As shown in Fig. 2(b), there are two gearboxes in the test
rig: a two-stage planetary gearbox and a two-stage fixed-axis gearbox. The present study just concerns the two-stage
planetary gearbox. In each of the two stages, an inner sun gear is surrounded by three or four rotating planet gears, and a
stationary outer ring gear. The load is transmitted through the sun gear to the planets, which ride on a planetary carrier.
The planetary carrier, in turn, transmits torque to the output shaft.
In a planetary gearbox, sun gear teeth easily suffer damage because their multiplicity of meshes with the planet gears
increases the potential for damage on the sun gear. In addition, the second stage of the planetary gearbox in our test rig
undertakes a larger load than the first stage. For these reasons, faults on the sun gear in the second stage planetary gearbox
are created in our experiments. The fault modes include a chipped tooth and a missing tooth, which are illustrated in Fig. 3.
There are 4 planet gears in the second stage of the planetary gearbox test rig, and 28, 36 and 100 teeth on the sun gear,
planet gears and ring gear, respectively.

Two-stage fixed-axis gearbox

NI data acquisition system Speed controller

Two-stage planetary gearbox

Brake

Laptop
Motor Tested gear

Planetary gearbox test rig

Fig. 2. A two-stage planetary gearbox test rig: (a) experimental system and (b) simulated model.

The chipped tooth The missing tooth

Fig. 3. Damaged sun gears: (a) having a chipped tooth and (b) having a missing tooth.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
4 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

Since vibration-based analysis is one of the principal tools for diagnosing mechanical faults [27–28], the vibration is
measured for a normal gear, the gears with a chipped tooth and a missing tooth using a tri-axial accelerometer. The
accelerometer was produced by PCB Electronics with the model number 356A12. It is mounted on the planetary gearbox
casing. A NI data acquisition system and a laptop with the data acquisition software are used to collect the vibration data
for further processing. The speed of the motor and the load of the magnetic brake are varied to simulate the general
planetary gearbox operating conditions. The vibration signals are acquired under the loaded condition and two different
motor speeds (2400 and 3000 rpm). The load applied to the gears equals 13.5 Nm in our experiments. The sampling
frequency is 5120 Hz. Since the vertical vibrations are constrained by the basement and the gravity of the test rig, their
amplitudes are not as large as those of the horizontal vibrations. Accordingly, the vibration signals of the horizontal
direction are more sensitive to the damage and therefore they are considered and analyzed in this paper. In the planetary
gearbox test rig, multiple meshing pairs of planet/sun and planet/ring produce vibration and the vibration signals
measured by the accelerometer come from all of the interactions after they propagate through the complex structure. Thus
it is extremely challenging to discover the weak characteristics of faults in the planetary gearbox.

3. Adaptive stochastic resonance (ASR) method

3.1. Theory of stochastic resonance

There are three necessary ingredients for establishing a SR system: (1) a bistable or multistable system (2) a weak
coherent input (a periodic or aperiodic signal) and (3) a source of noise that is inherent in the system or that added to the
coherent input [24]. The commonly adopted model for investigating SR is a bistable system, which is described by the
nonlinear Langevin equation as follows [24,29,30]:
dx 0
¼ U ðxÞ þ MIðtÞ þ C xðtÞ ð1Þ
dt
where x represents the displacement of a Brownian particle, I(t)¼S(t)þN(t) is an input signal embedded in a noisy
environment with signal S(t) and inherent noise N(t), and x(t)denotes the external noise added to induce SR and is
generally a zero-mean and Gaussian white noise subject to
E½xðtÞxðt þ tÞ ¼ 2DdðtÞ ð2Þ
in which D is the noise intensity and E [  ] stands for the operation for calculating the statistical mean value. M and C are
adjustable parameters. U(x) given by Eq. (3) denotes the symmetric double-well potential with system parameters a and b
characterizing the system
a b
UðxÞ ¼  x2 þ x4 ð3Þ
2 4 ˙
pffiffiffiffiffiffiffiffi
The symmetric double wells show that the minima locate at 7xm, where xm ¼ a=b. A potential barrier separates the
2
two minima with the height given by DU ¼a /4b. The barrier top locates at xb ¼ 0. Once the input signal, noise and
nonlinear system cooperate well, the potential barrier can be reduced and the particle, which stays in one potential well,
may surmount the energy barrier and enter another potential well. That is to say that the SR phenomenon happens. As a
result, the intensity of the characteristic signal increases and the detection of weak characteristics embedded in noisy
background will be possible. The output signal of the SR system may be obtained with a higher signal-to-noise ratio in
comparison with the input signal.
From the above analysis, it is seen that the height of the potential barrier DU is the critical factor deciding the
occurrence of the SR phenomenon, while DU is jointly decided by system parameters a and b. Thus, whether the SR
phenomenon will happen or not definitely depends on a and b. For an input signal, it is believed that there is always a pair
of system parameters a and b being able to induce the SR phenomenon well. Therefore, how to set and optimize the best
combination of system parameters a and b is an important issue in SR.

3.2. The ASR method

In this paper, an adaptive stochastic resonance (ASR) method is proposed by using ant colony algorithms to
automatically search and optimize SR parameters a and b. The ant colony algorithms are one kind of optimization
algorithms inspired by the foraging behavior of real ants. When searching for food, ants initially explore the area
surrounding their nest in a random manner. As soon as an ant finds a food source, it evaluates the quantity and quality of
the food and carries some of it back to the nest. During the return trip, the ant deposits a chemical pheromone trail on the
ground. The quantity of pheromone deposited, which may depend on the quantity and quality of the food, will guide other
ants to the food source. The indirect communication between the ants via pheromone trails enables them to find the
shortest paths between the nest and food sources [31]. This characteristic of real ant colonies is exploited in artificial ant
colonies in order to solve difficult combinatorial optimization problems such as the optimization of the best combination
of system parameters a and b in this study. Using the ant colony algorithms, the optimal parameters a and b can be
obtained to adaptively implement the SR system best matching input signals. Hence, the weak characteristics of the input

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]] 5

signals embedded in heavy noise may be discovered and the faults of planetary gearboxes can be detected using the ASR
method.
It is well known that SR requires input signal I(t)with small parameters, i.e., amplitude Ao o1 and frequency fo o1 of
periodic signal S(t) in I(t) and intensity D o o1 of external noise x(t). The vibration signals collected in the field of
machinery fault diagnosis, however, generally have much larger values of amplitude and/or frequency and/or noise
intensity than 1. Thus, to meet the requirement of small parameters, the input signals should be subject to necessary
preprocessing techniques before being input into the SR method. The currently investigated preprocessing techniques for
SR include scale normalization, modulation, re-scaling frequency, etc. In the proposed ASR method, frequency-shifted and
re-scaling technique, introduced and demonstrated in Ref. [30], is adopted to preprocess the input signals. Then the ant
colony algorithms are used to search the best combination of system parameters a and b, in which the initialization of the
variables can be referred to Ref. [31]. The signal-to-noise ratio of the output signal of SR is selected as the objective
function in the optimization process of ant colony algorithms.
The flow chart of the ASR method is shown in Fig. 4. It consists of the following procedural steps. (1) The variables of the
ant colony algorithms are initialized and the objective function is established based on the signal-to-noise ratio. (2) Using
the frequency-shifted and re-scaling technique, the vibration signals collected from the planetary gearbox test rig are
preprocessed to meet the requirement of small parameters of SR. (3) The ant colony algorithms are employed to adaptively
select and optimize SR parameters a and b. (4) In terms of the objective function, different combinations of parameters a
and b are compared, and the best one is saved. (5) The best combination is assigned to SR and the ASR system is developed.
(6) Using the ASR system, the weak characteristics immersed in the vibration signals of the planetary gearbox are
highlighted and detected. (7) The faults occurring in the planetary gearbox are eventually diagnosed according to the
detected characteristics.
To illustrate the usefulness of the ASR method in extracting weak characteristics, we generate a simulation signal with
the length of 2000 sampling data points, which is a sine wave embedded in heavy noise. The amplitude A and the
frequency f of the sine wave are 0.2 and 65 Hz respectively, and the intensity D of the noise equals 2. Therefore, the noise is
much stronger than the sine wave. Fig. 5(a) and (b) shows the time-domain waveform and the frequency spectrum of the
simulation signal, respectively. In both the original signal and the spectrum, the weak characteristic frequency of 65 Hz is
completely swamped by the heavy noise. Following the steps of the ASR method described in Fig. 4, we obtain the output
signal and its frequency spectrum given in Fig. 6. Although the sine wave cannot be seen from the output signal, its
frequency is clearly detected in the spectrum. The computation times of the ASR method are about 37 s for processing this
simulation signal.
For comparison, the EMD method, one of the well-studied signal processing techniques, is also used to decompose the
simulation signal. EMD is based on the local characteristic time scales of a signal and could decompose the signal into a set

Start

Initialize the variables and set the objective function

Preprocess the signals using the frequency-shifted and re-scaling technique

Optimize the SR parameters with the ant colony algorithms

Compare and record the optimal SR parameters

Construct the ASR system based on the optimal parameters

Discover the weak characteristics embedding in the signals

Diagnose faults of the planetary gearboxes

End

Fig. 4. Flow chart of the ASR method.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
6 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

Amplitude (mms-1)
Amplitude (mms-1)
10 0.4

0 0.2

-10 0
0 1 2 3 4 0 50 100 150 200 250
Time (s) Frequency (Hz)

Fig. 5. Simulation signal: (a) time-domain waveform and (b) frequency spectrum.
Amplitude (mms-1)

Amplitude (mms-1)
5 1
Characteristic frequency 65Hz
0 0.5

-5 0
0 1 2 3 4 0 50 100 150 200 250
Time (s) Frequency (Hz)

Fig. 6. Results of the ASR method for the simulation signal: (a) output signal and (b) frequency spectrum of the output signal.

5 0.4
0 0.2
-5 0
0 1 2 3 4 0 50 100 150 200 250
5 0.2
0 0.1
-5 0
0 1 2 3 4 0 50 100 150 200 250
5 0.2
0 0.1
Amplitude (mms-1)

0
Amplitude (mms-1)

-5
0 1 2 3 4 0 50 100 150 200 250
2 0.2
0 0.1
-2 0
0 1 2 3 4 0 50 100 150 200 250
2 0.2
0 0.1
-2 0
0 1 2 3 4 0 50 100 150 200 250
1 0.4
0 0.2
-1 0
0 1 2 3 4 0 50 100 150 200 250
2 0.2
0 0.1
-2 0
0 1 2 3 4 0 50 100 150 200 250
Time (s) Frequency (Hz)

Fig. 7. Results of the EMD method for the simulation signal: (a) IMFs and (b) frequency spectra of the IMFs.

of complete and almost orthogonal components named intrinsic mode function (IMF). The IMFs indicate the natural
oscillatory mode imbedded in the signal [32]. The IMFs decomposed by EMD are presented in Fig. 7(a) and (b) displays the
spectra of the IMFs. Observing each IMF and its spectrum in, we can discover neither the sine wave nor the characteristic
frequency 65 Hz. These results indicate that the EMD method fails to detect the weak characteristics embedded in
heavy noise. In other words, the ASR method provides obvious advantages over the EMD method in extracting weak
characteristics from the signal with a low signal-to-noise ratio.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]] 7

4. Diagnosis results of the planetary gearbox and discussions

4.1. Diagnosis results

The ASR method is applied to diagnosing faults of the planetary gearbox in this section. As introduced in Section 2, faults
of a chipped tooth and a missing tooth are simulated on the sun gear in the second stage of a planetary gearbox test rig. The
vibration signals are collected from the test rig under the loaded condition and two different motor speeds. The motor speeds
and the characteristic frequencies of the second stage planetary gearbox are concerned and summarized in Table 1.
The vibration signals collected under the motor speed of 3000 rpm are plotted in Fig. 8(a)–(c), which represents three
cases of the normal gear, the gears with a chipped tooth and with a missing tooth, respectively. It is observed from the
time-domain waveforms of the vibration signals that the two damaged cases produce a little larger vibration than the
normal one, and some impulses embedded in the vibration signals. However, these characteristics are masked by noise
and not evident enough to detect the existence of faults.
After analyzing the frequency spectra of the three cases, we find that the meshing frequency and its harmonics and the
rotating frequency and its harmonics of the sun gear in the first stage planetary gearbox are quite dominant. But the
characteristic frequencies of the second stage planetary gearbox cannot be seen clearly. The fact, however, is that the faults
we created locate in the second stage. Thus, we zoom in the frequency spectra with respect to the meshing frequency of
the second stage and present the zoomed-in spectra in Fig. 8(d)–(f). The characteristic frequencies, i.e. meshing frequency
and sidebands, of the second stage planetary gearbox are still vague in the zoomed-in spectra. This implies that it is
unfeasible to detect the fault characteristics of the sun gear of the second stage from both the vibration signals and their
frequency spectra.

Table 1
Motor speeds and characteristic frequencies of the second stage planetary gearbox.

Motor speed (rpm) fmotor (Hz) fs (Hz) fp (Hz) fr (Hz) fc (Hz) fp–p (Hz) fmesh (Hz)

3000.00 50 8.333 5.064 0.00 1.823 7.292 182.292

2400.00 40 6.667 4.051 0.00 1.458 5.833 145.833

Note: fmotor, fs, fp, fr and fc are the rotating frequencies of the motor, sun gear, planet gears, ring gear and carrier, respectively. fp–p and fmesh are the pass
frequency of the planet gears and the meshing frequency of the second stage, respectively.

x 10-3
Acceleration (g)

2
5
4fmotor
0

-2
0
0 1 2 3 4 170 180 190 200 210 220

x 10-3
Acceleration (g)

2
5

-2
0
0 1 2 3 4 170 180 190 200 210 220

x 10-3
Acceleration (g)

2
5
4fmotor
0

-2
0
0 1 2 3 4 170 180 190 200 210 220
Time (s) Frequency (Hz)

Fig. 8. Original vibration signals: (a) normal (b) a chipped tooth and (c) a missing tooth; zoomed-in spectra: (d) normal (e) a chipped tooth and (f) a
missing tooth.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
8 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

Then the proposed ASR method is applied to processing the above three vibrations signals. As mentioned in Section 3.2,
the signals are preprocessed using the frequency-shifted and re-scaling technique first and then input into the SR system
with its parameters adaptively optimized by the ant colony algorithms. Corresponding to the three health conditions of the
sun gear, the ASR method produces three output signals displayed in Fig. 9(a)–(c). Fig. 9(d)–(f) shows their whole spectra
of the three output signals. The computation times of the ASR method are about 305 s for processing each vibration signal.
Similar to the original vibration signals, the time-domain waveforms of the output signals cannot provide useful
information for diagnosing the gear faults either. To clearly show the frequency components, the whole spectra in
Fig. 9(d)–(f) are zoomed and the zoomed-in spectra are presented in Fig. 10(a)–(c). Key characteristics indicating the
occurrence of gear faults in the planetary gearbox are revealed in the frequency spectra of the output signals shown in
Fig. 10(a)–(c).
Comparing the frequency spectra shown in Fig. 10(a)–(c) with those of Fig. 8(d)–(f), it is found that the meshing
frequency and the sidebands of the second stage planetary gearbox are evidently discovered for the two damaged cases.
These characteristic frequencies are indicating damages occurring in the second stage planetary gearbox. Moreover, the
Acceleration (g)

0.04
0.2

0 0.02
-0.2
0
0 1 2 3 4 0 500 1000 1500 2000 2500
Acceleration (g)

0.04
0.2

0 0.02
-0.2
0
0 1 2 3 4 0 500 1000 1500 2000 2500
Acceleration (g)

0.04
0.2
0 0.02
-0.2
0
0 1 2 3 4 0 500 1000 1500 2000 2500
Time (s) Frequency (Hz)

Fig. 9. Output signals of the ASR method: (a) normal (b) a chipped tooth and (c) a missing tooth; whole spectra of the output signal: (d) normal (e) a
chipped tooth and (f) a missing tooth.

0.04 0.04
Acceleration (g)

Sidebands
fmesh
0.02 0.02

0 0
170 180 190 200 210 220 170 180 190 200 210 220
Frequency (Hz)

0.04
Acceleration (g)

Sidebands

0.02 fmesh

0
170 180 190 200 210 220
Frequency (Hz)

Fig. 10. Zoomed-in spectra of the ASR method: (a) normal (b) a chipped tooth and (c) a missing tooth.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]] 9
Acceleration (g)

2 2

0 0

-2 -2
0 1 2 3 4 0 1 2 3 4
Time (s) Time (s)

x 10-3 x 10-3
Acceleration (g)

5
4fmotor 5 4fmotor

0 0
130 140 150 160 170 180 130 140 150 160 170 180
Frequency (Hz) Frequency (Hz)

0.04
Acceleration (g)

Sidebands
0.04
Sidebands
fmesh
0.02 0.02 fmesh

0 0
130 140 150 160 170 180 130 140 150 160 170 180
Frequency (Hz) Frequency (Hz)

Fig. 11. Chipped tooth: (a) Vibration signal (b) original spectrum and (c) spectrum of the ASR method; missing tooth (d) vibration signal (e) original
spectrum and (f) spectrum of the ASR method.

value of the sidebands is almost equal to the rotating frequency of the sun gear, which suggests the damage on the sun
gear in the second stage planetary gearbox. This diagnosis result matches the fact that the faults of a chipped tooth and a
missing tooth are created on the sun gear in the second stage. Thus, the ASR method is able to detect the faults in the
planetary gearbox effectively.
To test the robustness of the ASR method, it is also applied to processing the vibration signals collected under the motor
speed of 2400 rpm. Fig. 11(a) and (d) illustrate the time-domain waveforms of the vibration signals for the gears with a
chipped tooth and a missing tooth. Fig. 11(b) and (e) shows their frequency spectra. The frequency spectra of the output
signals generated by the ASR method are displayed in Fig. 11(c) and (f). For the vibration signal colleted under a lower
rotating speed, it is generally more difficult to extract the fault characteristic frequencies from it. Thus, the meshing
frequency and the sidebands shown in Fig. 11(c) and (f) are not as evident as those of the motor speed of 3000 rpm shown
in Fig. 10 (b) and (c). However, Fig. 11 (c) and (f) still reveal obvious fault characteristic frequencies compared with the
original spectra shown in Fig. 11(b) and (e). These characteristic frequencies are sufficient to provide a means of detecting
the faults on the sun gear. Hence, the diagnosis result for the low motor speed confirms the effectiveness of the proposed
ASR method in diagnosing faults in the planetary gearbox.
The EMD method is also utilized to decompose the above vibration signals collected under the motor speed of
2400 rpm for comparisons. Figs. 12 and 13 present the processing results for a chipped tooth and a missing tooth
respectively. By examining all IMFs and the spectra in Figs. 12 and 13, we cannot see the characteristic frequencies of gear
faults for both two damage modes at all. Thus, it is demonstrated that the proposed ASR method is superior to the EMD
method in discovering weak characteristics for planetary gearbox fault diagnosis.

4.2. Discussions

1) A planetary gearbox is more complicated than a fixed-axis gearbox. It is a compound assembly of various parts that
constantly interact. Accordingly the vibration transmission paths from gear meshing points to transducers are
complicated. The complicated transmission paths may deteriorate or attenuate vibration response of faulty gears
through dissipation and interference effects, though the gear fault is quite severe (a missing tooth). This point can be
verified by contrasting the damaged gears displayed in Fig. 3 and the vibration signals and their spectra shown in Fig. 8,
from which both cyclic impacts and characteristic frequencies cannot be observed at all. It is concluded that the

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
10 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

5 0.1
0 0.05
-5 0
0 1 2 3 4 0 500 1000 1500 2000 2500
2 0.1
0 0.05
-2 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.5 0.01
0 0.005
-0.5 0
Acceleration (g)

0 1 2 3 4 0 500 1000 1500 2000 2500

0.2 0.04
0 0.02
-0.2 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.1 0.02
0 0.01
-0.1 0
0 1 2 3 4 0 -3 500 1000 1500 2000 2500
0.1 x 10
5
0
-0.1 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.05 x 10 -3
5
0
-0.05 0
0 1 2 3 4 0 500 1000 1500 2000 2500
Time (s) Frequency (Hz)

Fig. 12. Results of the EMD method for a chipped tooth: (a) IMFs and (b) frequency spectra of the IMFs.

2 0.1
0 0.05
-2 0
0 1 2 3 4 0 500 1000 1500 2000 2500
2 0.1
0 0.05
-2 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.5 0.02
0 0.01
-0.5 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.5 0.04
Acceleration (g)

0 0.02
-0.5 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.2 0.01
0 0.005
-0.2 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.1 0.01
0 0.005
-0.1 0
0 1 2 3 4 0 500 1000 1500 2000 2500
0.05
x 10-3
0 4
2
-0.05 0
0 1 2 3 4 0 500 1000 1500 2000 2500
Time (s) Frequency (Hz)

Fig. 13. Results of the EMD method for a missing tooth: (a) IMFs and (b) frequency spectra of the IMFs.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]] 11

vibration signals depend on not only damage itself but also the machine system. In addition, maybe because the second
stage planetary gear-pair works under bigger load and smaller speed than the first stage, the cyclic impacts of the
second stage are constrained in time domain and are not excited as large as those of the first stage. Consequently, the
fault characteristics cannot be detected based on the frequency spectrum. That is why we presented the ASR method to
extract fault characteristics of the second stage planetary gear-pair in this paper.
2) In the present study, the ASR method is developed to deal with the challenging issue of weak feature extraction in fault
diagnosis of planetary gearboxes. From the analyses and comparisons, it is appropriate to draw the following
conclusions that the proposed ASR method can extract fault characteristics of planetary gearboxes under different
motor speeds. The success obtained by the proposed method can be mainly attributed to the powerful property in weak
characteristic detection of SR which can utilize noise to enhance signal weak characteristics effectively.
3) However, the key components in a planetary gearbox include not only sun gear but also planet gears, ring gear and
bearings, etc. It is definitely true that defects on the planet gears are harder to be detected than those on the sun gear
because of their complex motion of not only rotating around their own centers but also revolving around the center of
the sun gear, and meshing simultaneously with the sun gear and the ring gear. The authors are conducting experiments
on the planetary gearbox test rig with planet gear faults and will address the fault diagnosis issues of planet gears in
their future work.

5. Conclusions

In this paper, different behaviors of planetary gearboxes from fixed-axis ones are pointed out first. Then an adaptive
stochastic resonance (ASR) method is proposed to solve the problem of weak feature extraction in fault diagnosis of
planetary gearboxes. In the proposed method, ant colony algorithms are utilized to synchronously select and optimize
stochastic resonance parameters. Therefore the optimal stochastic resonance system matching input signals can be
adaptively implemented. It is expected that weak characteristics can be effectively detected and planetary gearbox faults
diagnosed using the proposed method. A planetary gearbox test rig is established and two modes of faults in the test rig
are simulated. The vibration signals collected under different motor speeds are used to test the effectiveness of the
proposed method. The experimental results show that the method enables the detection of abnormalities in the planetary
gearbox and at the same time the identification of the fault severity degree.

Acknowledgments

This research is supported by National Natural Science Foundation of China (51005172), New Century Excellent Talents
in University (NCET-11–0421), Visiting Scholar Foundation of the State Key Laboratory of Mechanical Transmission in
Chongqing University (SKLMT-KFKT-201102) and Fundamental Research Funds for the Central Universities.

References

[1] J. McNames, Fourier series analysis of epicyclic gearbox vibration, ASME J. Vib. Acous. 124 (2002) 150–152.
[2] F. Chaari, T. Fakhfakh, M. Haddar, Dynamic analysis of a planetary gear failure caused by tooth pitting and cracking, J. Failure Anal. Prevention 6 (2)
(2006) 73–78.
[3] J.A. Hines, D.S. Muench, J.A. Keller, et al., Effects of time-synchronous averaging implementations on HUMS features for UH-60A planetary carrier
cracking, in: Proceedings of the American Helicopter Society International 61st Annual Forum, Grapevine Texas, USA, 2005, 218–227.
[4] D.M. Blunt, J.A. Keller, Detection of a fatigue crack in a UH-60A planet gear carrier using vibration analysis, Mech. Syst. Signal Process. 20 (8) (2006)
2095–2111.
[5] M. Mosher, Understanding vibration spectra of planetary gear systems for fault detection, Proceedings of the ASME Design Engineering Technical
Conferences, Paper no. PTG-48082, IL, USA, September 2003.
[6] P.D. McFadden, J.D. Smith, An explanation for the asymmetry of the modulation sidebands about the tooth meshing frequency in epicyclic gear
vibration, Proceedings of the Institution of Mechanical Engineers 199 (1985) 65–70.
[7] P.D. McFadden, A technique for calculating the time domain averages of the vibration of the individual planet gears and the sun gear in an epicyclic
gearbox, J. Sound Vib. 144 (1) (1991) 163–172.
[8] W. Bartelmus, R. Zimroz, Vibration condition monitoring of planetary gearbox under varying external load, Mech. Syst. Signal Process. 23 (2009)
246–257.
[9] W. Bartelmus, R. Zimroz, A new feature for monitoring the condition of gearboxes in non-stationary operating conditions, Mech. Syst. Signal Process.
23 (2009) 1528–1534.
[10] Y.G. Lei, D.T. Kong, J. Lin, et al., Fault detection of planetary gearboxes using new diagnostic parameters, Meas. Sci. Technol. 23 (5) (2012) 1–10.
[11] Y.G. Lei, J. Lin, Z.J. He, et al., A method based on multi-sensor data fusion for fault detection of planetary gearboxes, Sensors 12 (2) (2012) 2005–2017.
[12] T. Barszcz, R.B. Randall, Application of spectral kurtosis for detection of a tooth crack in the planetary gear of a wind turbine, Mech. Syst. Signal
Process. 23 (2009) 1352–1365.
[13] B. Zhang, T. Khawaja, R. Patrick, et al., Application of blind deconvolution denoising in failure prognosis, IEEE Trans. Instrum. Meas. 58 (2) (2009)
303–310.
[14] B. Zhang, T. Khawaja, R. Patrick, et al., A novel blind deconvolution de-noising scheme in failure prognosis, Trans. Inst. Meas. Control 32 (1) (2010)
3–30.
[15] P.D. Samuel, D.J. Pines, A review of vibration-based techniques for helicopter transmission diagnostics, J. Sound Vib. 282 (2) (2005) 475–508.
[16] P.D. Samuel, D.J. Pines, Constrained adaptive lifting and the CAL4 metric for helicopter transmission diagnostics, J. Sound Vib. 319 (2) (2009)
698–718.
[17] R. Zimroz, W. Bartelmus, Gearbox condition estimation using cyclo-stationary properties of vibration signal, Key Eng. Mater. 413–414 (2009)
471–478.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021
12 Y. Lei et al. / Mechanical Systems and Signal Processing ] (]]]]) ]]]–]]]

[18] A. Bartkowiak, R. Zimroz, Outliers analysis and one class classification approach for planetary gearbox diagnosis, J. Phys.: Conference Series 305
(2011) 012031.
[19] A. Bartkowiak, R. Zimroz, Data dimension reduction and visualization with application to multidimensional gearbox diagnostics data: comparison of
several methods, Solid State Phenom. 180 (2012) 177–184.
[20] M. Inalpolat, A. Kahraman, A theoretical and experimental investigation of modulation sidebands of planetary gear sets, J. Sound Vib. 323 (4) (2009)
677–696.
[21] M. Inalpolat, A. Kahraman, A dynamic model to predict modulation sidebands of a planetary gear set having manufacturing errors, J. Sound Vib. 329
(4) (2010) 371–393.
[22] W.D. Mark, J.A. Hines, Stationary transducer response to planetary-gear vibration excitation with non-uniform planet loading, Mech. Syst. Signal
Process. 23 (4) (2009) 1366–1381.
[23] W.D. Mark, Stationary transducer response to planetary-gear vibration excitation II: effects of torque modulations, Mech. Syst. Signal Process. 23 (7)
(2009) 2253–2259.
[24] R. Benzi, A. Sutera, A. Vulpiana, The mechanism of stochastic resonance, Physics A 14 (1981) 453–457.
[25] L. Gammaitoni, P. Hanggi, P. Jung, et al., Stochastic resonance, Rev. Modern Phys. 70 (1998) 223–287.
[26] F. Chapean-Blondeau, Stochastic resonance at phase noise in signal transmission, Phys. Rev. E 61 (2000) 940–943.
[27] Y.G. Lei, Z.J. He, Y.Y. Zi, et al., New clustering algorithm-based fault diagnosis using compensation distance evaluation technique, Mech. Syst. Signal
Process. 22 (2008) 419–435.
[28] Y.G. Lei, Z.J. He, Y.Y. Zi, et al., Fault diagnosis of rotating machinery based on multiple ANFIS combination with GAs, Mech. Syst. Signal Process. 21
(2007) 2280–2294.
[29] S.F. Xie, B.R. Xiang, H.S. Deng, et al., Improved stochastic resonance algorithm for enhancement of signal-to-noise ratio of high-performance liquid
chromatographic signal, Anal. Chim. Acta 585 (2007) 60–65.
[30] J.Y. Tan, X.F. Chen, J.Y. Wang, et al., Study of frequency-shifted and re-scaling stochastic resonance and its application to fault diagnosis, Mech. Syst.
Signal Process. 23 (2009) 811–822.
[31] M. Dorigo, C. Blum, Ant colony optimization theory: a survey, Theor. Comput. Sci. 344 (2005) 243–278.
[32] N.E. Huang, Z. Shen, S.R. Long, et al., The empirical mode decomposition and the Hilbert spectrum for nonlinear and non–stationary time series
analysis, Proceedings of the Royal Society of London A 454 (1998) 903–995.

Please cite this article as: Y. Lei, et al., Planetary gearbox fault diagnosis using an adaptive stochastic resonance method,
Mech. Syst. Signal Process. (2012), http://dx.doi.org/10.1016/j.ymssp.2012.06.021