Professional Documents
Culture Documents
1 Introduction
Gears are key components of modern power transmission systems in different
industrial applications. Particularly in case of high power density and depending
on operating conditions, different damage modes such as micro-pitting, pitting,
or scuffing are observed on the tooth flanks of gears. Therefore, the interest in
the development of monitoring systems for online diagnosis and prediction of
different machine states, which enable condition-based or predictive mainte-
nance, grew recently.
Condition-based maintenance (CBM) has been developed as a maintenance
policy, which contributes to the improvement of reliability and availability of
technical systems through maintenance decisions based on real-time infor-
mation regarding the current health state of a system. This multidisciplinary field
of research comprises two major aspects, namely diagnostics and prognostics
[1]. To detect damages present in structures or systems (diagnosis) and to ob-
tain a prediction of the remaining useful life (RUL) (prognosis), the data ac-
quired during continuous online monitoring of technical systems are used along
(a) (b)
Figure 1: Examples of micro-pitting (a) and pitting (b) faults
© Chair of Industrial and Automotive Drivetrains, Ruhr-Universität Bochum
loading causes these cracks to propagate in the material until finally reaching
the surface forming macroscopic craters due to extensive loss of material.
Recently, Acoustic Emission (AE) technique gained attention in the context of
gear transmission monitoring for detection of different damages, i.e. pitting [5] or
tooth root crack [4]. In this regard, asperity contact was identified as major
source of AE in meshing gears [6, 7]. Here, mixed AE waveforms are observed,
where burst type and continuous AE are presumably related to rolling and slid-
ing contact of asperities, respectively [6]. Furthermore, the AE related to mesh-
ing gears reportedly masks possible bearing faults [8].
Signal processing of AE typically involves the extraction of time-domain features
(i.e. count rate, peak amplitude or rise time) which are extracted with respect to
a predefined threshold value to characterize AE measurements. However, this
method is not viable for processing of continuous AE signals. Therefore, statis-
tical features of AE waveforms including Root Mean Square (RMS) [9], Kurtosis
(KUR) [4] or Crest Factor (CF) [7] are frequently used in context of gears to
identify damage-specific features of different gearbox damages. However, these
statistical features usually provide only the distinction of two different states in a
fault/no fault manner. For instance, Samanta [10] discriminated two different
states (healthy and pitting fault) of a two-stage gearbox using 45 statistical fea-
tures which were extracted from vibration measurements using SVM and ANN
for classification. To implement a damage detection strategy capable to discrim-
inate multiple damage modes advanced feature extraction techniques are re-
quired. Li et al. [11] used different features from the time and time-frequency
domain and proposed a feature selection scheme based on Genetic Algorithm
(GA) for damage detection in a single-stage gearbox. Using the proposed
scheme, good performances could be achieved on the classification of 8 differ-
ent gearbox faults using different classification algorithms.
Experimental setup
Measurements were performed on a standard gearbox test rig, which is usually
used in connection with standardized procedures for wear-testing of lubricated
gears. Originally, this test rig was developed at Forschungsstelle für Zahnräder
und Getriebebau (FZG). A schematic drawing of the FZG test rig is shown in
Figure 2. The test rig contains two cylindrical gear stages, namely testing and
slave gears in a back-to-back arrangement. The slave gears are driven by an
electric drive. Static torque loading of the gears is achieved via a torque clutch.
Defined loads, which are specified in terms of load stages (LS), can be applied
by means of calibrated weights and a lever arm. Thus, the electric drive only
provides energy losses due to friction resulting from the gears and bearings ac-
cording to the principle of circulating power. During the experiments, similar
pairs of gears in different condition, including normal operation, micro-pitting
and pitting, are used. The pinion of each pair has 16 teeth and meshes with a
wheel with 24 teeth providing a transmission ratio of i = -1.5. The wheel is driv-
en at a speed of 1450 rpm providing rotational speed of 2175 rpm at the pinion
shaft, which corresponds to a gear meshing frequency of approximately 580 Hz.
In context of gearbox monitoring using AE, sensor placement is challenging. To
achieve low attenuation of the AE amplitudes, AE sensors are frequently
mounted directly on the gears. To this end, a slip-ring is most frequently used to
connect sensors attached to rotating parts to the acquisition hardware [3, 6, 7,
16–18]. To avoid attaching sensors to moving parts, AE sensors are frequently
mounted on fixed parts of gearbox bearings (i.e. bearing races) [17, 18]. Moreo-
ver, Loutas et al. [19] developed a custom mounting fixture for AE sensors
based on friction contact, where constant contact force between the fixed sen-
sor and the rotating wheel is achieved by a spring element. The mounting tech-
nique was verified using attenuation tests, where pencil-lead break tests were
performed at different locations. In this study, the AE sensor was mounted out-
side the housing of the gearbox in a horizontal orientation above the roller bear-
ing of the pinion, as indicated in Figure 2. Besides the interface between the AE
sensor and the housing, several interfaces are encountered along the propaga-
tion path including gear/shaft as well as shaft/bearing and bearing/housing inter-
Experimental results
In the sequel, results of AE measurements from a FZG gearbox test rig are pre-
sented. First, the sensitivity of the measurement chain to the meshing of the
gears is verified by comparing the measurement results obtained under different
operating conditions. Hereafter, joint time-frequency domain representation of
measurement results obtained from different machine states is studied in detail.
Here, different patterns are recognized depending on the wear state of the
gearbox and the causal relation to different damage modes is discussed.
In Figure 3, measurement results of the gearbox under normal operation and
with pitting fault are compared in time domain representation. In case of Fig-
ure 3 (a), the gears are fault free, which is considered the reference condition.
Here, the signal is characterized by transient waveforms, which show a periodic
pattern of similar peak amplitudes. Furthermore, the period duration of approxi-
mately 1.7 ms corresponds to a frequency of 588 Hz, which correlates with the
theoretical meshing frequency of the gears. Compared to this, the measurement
signal obtained from a pair of gears with pitting fault is presented in Figure 3 (b).
Visual inspection of the gears prior to the experiment revealed pitting at a single
tooth of the pinion, while the wheel was classified as fault free. Here, the pitting
Normalized Amplitude
1
0.5
0
−0.5
−1
0 5 10 15 20 25 30
Time [ms]
(a) Normal Operation
Normalized Amplitude
1
0.5
0
−0.5
−1
0 5 10 15 20 25 30
Time [ms]
(b) Pitting Fault
Figure 3: Time domain representation of AE measurement
defect acts as a marker indicating a full rotation of the pinion. From the meas-
urements, a similar periodic pattern is observed. However, compared to the
baseline pattern, additional discrete events are observed showing comparably
large peak amplitudes. In between each peak amplitude event, 16 periods of
the baseline pattern are observed.
From the above considerations it can be concluded that the measurement chain
is sensitive to AE which is related the gear meshing. Apparently, the observed
baseline pattern is related to the meshing of the fault free pinion teeth and the
wheel whereas peak amplitude events occur each time the faulty tooth enters
the mesh. Also, gear meshing seems to be dominant component of the meas-
urement signal compared to possible additional sources of AE, i.e. roller bear-
ings of the gearbox, which is consistent with literature [8].
In the following, the measurement results are considered in the joint time-
frequency domain. Here, different operating conditions are considered, includ-
ing normal operation (fault free) under different loads as well as micro-pitting
and pitting. Time-frequency domain analysis of the measurement signal is per-
formed by means of continuous wavelet transform. This method is particularly
suitable for the analysis of low amplitude, high frequency signals such as AE.
Due to variable time-frequency resolution, this transform provides improved
time resolution on increasing frequencies and thus enhanced interpretability.
Furthermore, rescaling of the obtained matrix of wavelet coefficients was per-
formed for illustrative purposes. Due to increasingly low signal intensities on
increasing frequencies, each row was normalized to a maximum value of unity.
Thus, decoupling of the color scales at different frequencies is achieved. This
procedure allows highlighting of low intensity effects at high frequencies.
In Figure 4 (a) characteristic patterns observed during normal operation are
presented. This is considered as reference condition. Besides a noise floor at
approximately 400 kHz, repetitively occurring peak frequencies are observed in
the spectrogram at frequencies between 40 kHz – 60 kHz. These observations
are similar among different loads. Considering Figure 4 (b), a different set of
gears is used. Prior to the experiments, visual inspection of the gears revealed
matte grey surface appearance of several tooth flanks indicating micro-pitting as
damage mode. Besides increased activity in the lower frequency range, a
downshift of the peak frequencies to approximately 30 kHz – 40 kHz is ob-
served. Furthermore, additional effects in the frequency range above 100 kHz
are apparent. In principle, increase in surface roughness leads to a reduction in
specific oil film thickness causing increase in asperity contacts [20]. Hence, en-
ergy in this frequency range is suspected to be related to an increase in surface
roughness indicating deteriorated surfaces of the tooth flanks. Presumably, AE
(c) Pitting
5 Bibliography
[1] A. K. S. Jardine, D. Lin, and D. Banjevic, “A review on machinery
diagnostics and prognostics implementing condition-based maintenance,”
Mech. Syst. Signal Process., vol. 20, no. 7, pp. 1483–1510, 2006.
[2] B. Lu, Y. Li, X. Wu, and Z. Yang, “A review of recent advances in wind
turbine condition monitoring and fault diagnosis,” in Power Electronics
and Machines in Wind Applications, 2009.
[3] B. Eftekharnejad and D. Mba, “Seeded fault detection on helical gears
with acoustic emission,” Appl. Acoust., vol. 70, no. 4, pp. 547–555, 2009.
[4] T. H. Loutas, G. Sotiriades, I. Kalaitzoglou, and V. Kostopoulos,
“Condition monitoring of a single-stage gearbox with artificially induced