Signal Processing to Reduce the Effect of Gear Dynamics

Eric Bechhoefer Yalin Ozturk

CEO/Chief Engineer Chief Engineer, Transmission Analysis
GPMS In Turkish Aerospace
Cornwall, VT Ankara, Turkey

ABSTRACT
An improved conditioning monitoring technique is provided for rotating components in gearboxes that account for
gear system dynamics. A rotation rate for the component is generated from vibration data by estimating the rotation
rate based on a tachometer measurement of another shaft and the shaft ratio. This estimated rotation rate is used,
together with the known configuration of the component, to estimate a known gear mesh frequency of the component.
By filtering for a range of frequencies around the gear mesh frequency based on variation in the shaft rate, the gear
mesh frequency can be determined, and from that signal, an actual rotation rate for the component can be determined.
The exact rate can then be used in deriving an analytic vibration spectrum for the component that is not degraded due
to gear system dynamics effects. Further, this allows the calculation of the gearbox transfer function.

INTRODUCTION 1 Condition monitoring based on vibration measurements

depends in part on the rotation rate of the monitored
Condition monitoring of components in machines is a component. The rotation rate of any component is a function
technique in which a measure of the health of components can of the measured (by a tachometer) shaft rate to any shaft in a
be realized by using sensors. Condition monitoring improves gearbox by the gear ratio between the measured shaft and the
asset reliability and allows for improved availability by shaft of the component of interest. The TSA itself is sensitive
allowing for opportunistic maintenance. Also, condition to noise, such as jitter (Ref 4), where it was shown that small
monitoring has been successful in improving safety. For changes in the calculated shaft rate due to jitter could
gearboxes, condition monitoring typically involves the use of adversely affect the TSA and the analysis resulting from it.
accelerometers to measure vibrational energies associated
with rotating components. The condition monitoring system
performs analysis of the accelerometer data to extract features
LIMITATIONS OF GEAR ANALYSIS
that are indicative of component damage and fatigue. It was recently observed that the measured rotation rate might
Vibration health monitoring systems may monitor different not result in a precise rotation rate of the component of
aspects of main gearbox components, such as gears, shafts, interest if the component of interest is displaced physically
and bearings, including shaft order 1 (first harmonic of the apart (in terms of gear dynamics) from the shaft that the
shaft under analysis), shaft order 2, gear mesh frequencies, tachometer is measuring. In other words, the actual rotation
gear tooth indicators, bearing wear indicators, and modulation rate can be affected by the phase error. This variation in actual
of web tone epicyclic gear indictors. versus measured rotational rates is due to the global effect of
gear system dynamics. It is deleterious to an analysis of a
The monitoring of gears and shafts is typically based on time-synchronous averaged vibration spectrum used for
analyses that use the Time Synchronous Average (TSA) of condition monitoring (e.g., a similar phenomenology as jitter
measured vibrational signals (Ref 1). Time synchronous on the tachometer signal). That is, if the tachometer is not
averaging is a method for reducing noise in spectra of mounted near the component (shaft or gear) under analysis,
complex signals as well as for extracting more information the analysis is degraded due to phase errors introduced by the
from the signals. The TSA may be used in many analyses, torsional dynamics of the system. The extent of this
including Residual/Different analysis, the Energy Operator, degradation depends on the dynamic characteristics of the
the Narrow Band Analysis, the Amplitude Modulation geared shaft system and the operational speed, i.e., the margin
Analysis, the Frequency Modulation Analysis, and other between the operational speed and related critical speeds
analyses (Ref 2, 3). (natural frequencies) in terms of torsional dynamics.

Presented at the Vertical Flight Society’s 76th Annual Forum &

Technology Display, Montreal, Quebec, Canada, May 19-21, 2020.
Copyright © 2020 by the Vertical Flight Society. All rights reserved.
1
An improved conditioning monitoring technique is presented This causes a smearing of the spectral content, as the energy
that accounts for gear system dynamics by determining the associated with a shaft or gear is spread across many
actual rotation rate for a component of interest from vibration frequencies. As can be seen in FIG.1, the shaft frequency
data. The determined rotation rate is, in this way, corrected varies from 534.2 Hz to 537.5 Hz. If one were interested in
for gear system dynamics and allows for improved signal the gear mesh frequency of this shaft (which has, as an
analysis of the component of interest so that faults or defects example, 35 teeth), then for this range of shaft rates, the gear
can be detected earlier and more accurately. The analytic mesh frequency would range from 18,697 Hz to 18,813 Hz.
signal is developed using an ideal filter in a single functional The sample rate of the condition monitoring system may be,
process, and noise is removed. This results in an improved for example, 93,750 samples per second. Using the FFT, the
TSA spectrum for rotating components in gearboxes, and length (i.e., the number of data points) of the analysis is
TSA spectra are fundamental to gear fault diagnostics using typically some radix-2 value, for example, 16,384. Given this
vibration-based analysis. window length, then each FFT bin would have 5.722 Hz of
frequency content. In this scenario, the frequency of the gear
GEAR ANALYSIS BASED ON THE TIME mesh is spread across 115 Hz or 20 bins. Hence, the
SYNCHRONOUS AVERAGE representation of a feature would not be accurate. This is one
of the reasons why gear and shaft analysis use the TSA, which
The time synchronous average (TSA) signal is generated by removes the effect of shaft speed change by resampling the
resampling data to account for variation in shaft rates, which data (Ref 1, 5).
confers many benefits. An assumption of the Fourier
transform is stationarity, i.e., that the signal does not change The TSA Algorithm
over time. Due to limits in the bandwidth of the motor/engine
controllers, there is always some change in the shaft rate. This Analysis of the TSA was well described in Ref 9., and perhaps
change in shaft rate can be measured directly with a first implemented for gearbox analysis in Ref 3 (R.M Stewart
tachometer, which is apparent from the data shown in Figure went on the develop perhaps the first HUMS). As per Ref 9.,
1, which shows the power turbine shaft rate of a light the function to be averaged, x(t), is digitized at a sample rate
helicopter over time. The tachometer sample rate is of nT, giving a waveform x(nT), with an averaged period of
48,000,000 samples per second and accurately measures the mT, such that:
speed of the engine power turbine shaft, and, as such, when
an appropriate ratio is applied, it can be used to find the shaft 𝑦(𝑛𝑇) = 1)𝑁 ∑!"#
$%& 𝑥(𝑛𝑇 − 𝑟𝑚𝑇) eq. 1
rates of other shafts in the gearbox. Note that the shaft rate is
not constant. The variation is due to changes in torque coming This is a linear filtering operation where it can be shown that
from the main, tail rotor, and the control cycle of the FADEC. the discrete transfer function is:
Any inputs from the flight control and in forwarded flight, the
added lift of the advancing blade, causes a torque ripple into 𝑌(𝑧) "'!
𝐻(𝑧) = 3𝑋(𝑧) = 1)𝑁 1 − 𝑧 )1 − 𝑧 "' eq. 2
the engine. Because of the finite bandwidth of the engine
controller, the engine speed changes, resulting in changes in
the shaft rate over time. This is a comb filter, where the filter response is based on the
number of averages, N, see Figure 2. Note that the x-axis is in
the order domain, where order 1 is the average, resampled
shaft rate, and higher-order are the corresponding harmonic.
Hence, for a 29-tooth gear, frequency content associated with
the gear mesh would be present at order 29. This frequency
not synchronous with the shaft would be filtered by,
effectively, a FIR filter of length N (the number of averages),
with coefficients bi = 1/N. The attenuation would be based on
the percent difference of the non-synchronous source. For
example, say the shaft under analysis had a rate of 40Hz and
was sampled for 16 revolutions. A nonsynchronous 44 Hz
source would then have an order of 1.1, which would be
attenuated by the TSA to 0.184 times the source level.

This clearly suggests that if the instantaneous frequency of,

say a gear mesh, is somewhat different than the measured
tachometer signal (due to jitter, or some gear dynamic
phenomenology), the gear mesh signal can become attenuated
Figure 1 Shaft rate derived from the tachometer by the TSA.
These changes in shaft rate over time invalidates an Practically speaking, the TSA is a simple algorithm. The TSA
assumption of stationarity when using the Fourier transform. resamples the vibration associated with a shaft or gear, in the
2
spatial domain, such that vibration associated with each shaft mesh damping constant, and e(t) is the external displacement
order, in the Fourier domain, represents one frequency bin. action, which is known as gear transmission error
For example, in a system in which the shaft rate is such that representing the gear profile errors and/or intended
for a given vibration sample rate, the acquisition system, on modifications on gear tooth profiles.
average collects 800 samples per revolution, the TSA would
resample the 800 samples to 1024 data points (1024 is the next Therefore, geared shaft systems have distinct dynamic
highest radix-2 value from 800). characteristics. Such systems, as noted, have time-varying
stiffness features, leading to a parametric excitation of the
dynamic system.

Figure 2 Filter response in the order domain, for N =

4, and 16
GEAR SYSTEM DYNAMICS
Gears often use an involute profile to transfer the power,
where the torque is transmitted through a conjugate action
between mating tooth pairs. Mesh stiffness, which can be
defined as the anti-deformation capacity of the gears under
contact, varies significantly during a mesh cycle. The primary
parameter which affects the change in mesh stiffness is the
variation of the number of tooth pairs in contact. In addition
to this, change in point of contact during a mesh cycle is also Figure 3 Simple lumped parameter 1DoF gear
essential, since it is accompanied by a continuous change in dynamics model
contact stiffness of the tooth pair and change in individual
bending stiffness of gears. A most significant mesh stiffness Hence, the torsional dynamics of the geared shaft systems are
variation is observed for spur gear pairs. It is an important not only affected by the shaft characteristics (i.e., geometry,
phenomenon for bevel gear dynamics as well. When such material, etc.) but also from the gear meshes and their distinct
systems are excited near the critical speeds, the dynamic dynamic features. Due to such complex dynamic behavior, it
response can even increase to the levels such that loss of is not always possible to be away from every critical speed of
contact or even double-sided contact may occur, which is an the overall dynamic system with a wide margin. This also
indication of strongly increased dynamic forces (Ref 6). Note means system dynamics can be crucial in vibration-based
that this parametric excitation may become dominant not only drive train health monitoring. Coupled dynamic modes can
at the fundamental mesh frequency but also at the harmonics also be existent where a dynamic phase difference is possible
of it due to the multi-harmonic content of the mesh stiffness. between any two rotating components of the system. This
means pseudo-tachometer readings, which are derived based
This behavior can be modeled as a damped mechanic on gear ratios between the shaft with a tachometer and geared
oscillator. A lumped-parameter, single degree of freedom shaft of interest, may not always provide precise results for
(SDOF) gear dynamics model is provided in Figure 3 for a TSA calculations.
gear pair. Here k(t) is the time-varying mesh stiffness, c is the

3
Such dynamic effects are suspected for the drive train under (i.e., addition) of many signals in the time domain. For
consideration in Figure 4, where the TSA spectral energy of a example, consider a simple gearbox with an input shaft, an
gear mesh is considerably less than the measured spectral output shaft, and a gear pair. The input shaft turns at 30 Hz
energy. and has a 32-tooth gear, and the output shaft has an 82-tooth
gear with a rotational speed of 11.707 Hz. The gear mesh
frequency is 960 Hz (30 * 32). The gear mesh frequency will
have sidebands as a result of any shaft imbalance being
modulated onto the gear mesh. This can be proved using the
trigonometric identity:

cos(𝑎) × 𝑐𝑜𝑠(𝑏) = 1)2 [cos(𝑎 + 𝑏) + 𝑐𝑜𝑠(𝑎 − 𝑏)] eq 3

where, in this example, cos(a) is 960 Hz, and cos(b) is 30 Hz

and/or 11.707 Hz shaft. Additionally, if the shaft is bent or
bowed, there will be a 2x shaft vibration component. Other
manufacturing defects, such as the gear not being mounted
perpendicular to the shaft, or not centering the shaft on the
gear (e.g., eccentricity), will result in additional frequency
tones.

To recover only those signals associated with the desired

component, one can use an ideal bandpass filter and create an
Figure 4 Raw vs. TSA spectrum analytic signal in one functional procedure. This is followed
When a local tachometer signal was calculated from the by using a jitter reduction model to remove noise (jitter) from
vibration data, it is seen that the “local” tachometer signal the reconstructed tachometer signal not associated with
shows a great deal of variation in rotation rate due to the changes in the machine rate.
dynamics of the system (Figure 5). In general, the phenomena
have not been observed because of limitations in the ability to Recovering rotational information from vibration data
determine the actual rotation rate of components involves using an estimate of the rotation rate of a component
“downstream” from the shaft being measured with a under analysis based on the tachometer measurement of a
tachometer. shaft and the shaft ratio to that component. From this
information, a known gear mesh frequency for the component
may be estimated. A range of frequency encompassing the
estimated gear mesh frequency is found based on the variance
in the measured shaft rate. By filtering around this range, the
actual gear mesh frequency may be extracted from the overall
vibration data even though its magnitude may be significantly
smaller than the average overall vibration spectrum. Once the
gear mesh frequency signal is determined, the actual shaft rate
of the component of interest can be found. This may be
accomplished, for example, by the following pseudo-code:

• Define the Sample Rate = sr. The number of data

points, n, of vibration data equals sr x acquisition
length in seconds, then:

• Calculate the next larger radix-2 length for the FFT:

nRadix = 2ceil(log2(n)).

Figure 5 Comparison of tachometer derived shaft From the tachometer measurement and the gearbox
rate vs. vibration derived shaft rate configuration (i.e., the shaft ratio from the shaft measured by
the tachometer to the shaft under analysis), calculate meta
Deriving the tachometer signal from the vibration
statistics such as approximate rotation rate of the shaft under
Vibration signals from rotating equipment are sinusoidal and, analysis (i.e., the first moment), the variation (i.e., the second
by definition, synchronous with signals associated with the moment) in the approximate rotation rate, and the estimated
shaft rotation. However, the nature of vibration requires known gear mesh frequency (also based on the number of
significant signal processing of vibration signals to extract teeth of gear on the shaft under analysis).
useful information. Measured vibration is the superposition

4
 • From the estimated known gear mesh frequency and A process may be used in which developing the analytic
the variance in the estimated shaft rotation rate, signal using an ideal bandpass filter is completed in a single
calculate the low bandwidth index and the high functional process. The analytic signal is defined for the real-
bandwidth index (bwlow, bwhigh), which will valued signal x(t), as determined:
encompass the gear mesh frequency of interest.
𝑋(𝑓) = 𝐹{𝑥(𝑡)} eq. 4
• Take the zero-padded FFT of the vibration data.
where F is the Fast Fourier Transform, and where:
• Zero the FFT from zero to bwlow, and from bwhigh
to nRadix. 𝑋( (𝑓) = 𝑋(𝑓), 𝑓 = 0 eq. 5

• Take the inverse FFT to generate the analytic signal. 𝑋( (𝑓) = 2𝑋(𝑓), 𝑓 > 0 eq. 6

• Calculate the unwrapped argument of the generated 𝑋( (𝑓) = 0, 𝑓 < 0 eq. 7

analytical signal from to 1 to n time series. (The
argument is the arctangent of the imaginary part of 𝑥( (𝑡) = 𝐹 "# K𝑋( (𝑓)L eq. 8
the analytic signal, to the real part of the analytic
signal. Note that the value can only go from 0 to p where X(f) is the Fourier transform of x(t), and f is measured
and -p to 0. One is interested in the cumulative signal frequency.
rotation of the analytic signal in time, hence as the
signal exceeds -p to some small positive number, 2p RESULTS
is added to it. That is if at index i the value is -0.03p,
By reconstructing the tachometer signal from the local gear
and the next value calculated at index i + 1 is 0.03p,
under analysis (Ref 7, 8), resulting zero-cross data improves
the saved (unwrapped) value is 2.03p.) the TSA, allowing for the full energy measured in the shaft
under analysis to be used for gear analysis.
• Normalize the time series of radians by the number
of teeth of the gear (assuming 1st order harmonics).

• Interpolate the number of indexes for every 2π

radians. The value 2π radians is one zero cross.
Hence, the interpolation gives the exact index of the
zero-cross of the shaft.

• Normalize to "tachometer" zero-crossing index by

the Sample Rate (sr), which provides the zero cross
time and from which the rotation rate for the
component under analysis is calculated.

Idealized bandpass filter

A bandpass filter is the convolution of a low pass filter with a
high pass filter. These filters are implemented as Finite
Impulse Response (FIR) filters to improve their stability.
However, even rather large filters have an inadequate
response. Consider a case of a wind turbine gearbox, with an Figure 6 Improved TSA spectrum from enhanced
approximately 29 Hz shaft and a 32-tooth gear, which has a analysis
frequency of 928 Hz. The bandwidth of the filter is set from Improving the signal to noise ratio using this technique will
910 Hz to 950 Hz to exclude the frequencies associated with enhance fault detection. This can be seen in Figure 6,
the 30 Hz of the high-speed shaft (i.e., 29 * 32 - 29 = 899 Hz comparing the raw spectrum and TSA spectrum (subplot 1),
and 29 * 32 + 29 = 957 Hz). Using a large, 120 tap FIR filter, to the reconstructed tachometer TSA in Figure 1, subplot 2,
the bandwidth (at 50% power, 3dB) of this filter is 1670 Hz. were the SNR gain can be seen to be almost 8 dB.
The filter does not reject the spectral content at 500 Hz, or
1121 Hz, because the filter is not ideal and suffers from This phenomenon has been observed for many gears in the
spectral leakage. These additional frequencies will be main transmission of helicopters. For example, Figure 7.
superimposed on the 928 Hz signal and will distort the desired shows the tachometer and reconstructed tachometer for the
analytic signal, reducing the quality of the resulting ring gear of a light single helicopter.
tachometer signal.

5
 𝑍𝐿𝑂𝐶𝐴𝐿(𝑓) = 𝐻(𝑡) × 𝑍(𝑡) eq. 10

Taking the log allows the nonlinear system to be expressed as

a linear system. This technique is commonly known as
homomorphic deconvolution.

log{𝑍𝐿𝑂𝐶𝐴𝐿(𝑓)} − 𝑙𝑜𝑔{𝑍(𝑡)} = 𝑙𝑜𝑔{𝐻(𝑡)} eq. 11

The transform function, for example, 1 (taken from the input

and output of Figure 4), is shown in Figure 9.

Figure 7 Tachometer and reconstructed tachometer

signal, main gearbox
Note that because of the relatively high gear ratio of the main
gearbox and resulting low speed of the main rotor, there are
only 34 revolutions over the five-second acquisition. A
comparison of spectrum using the TSA, vs. the "local"
tachometer derived from the vibration signal, Figure 8 shows
a 4-dB improvement in the signal then on the figure itself.

Figure 9 The measured transfer function due to gear

dynamics
As seen from the phase change, the gearbox goes through
resonance at approximately 17 Hz, as indicated by the phase
change.

CONCLUSIONS
It has been observed in the TSA of some main gearbox
analysis that the TSA spectrum had less energy than the raw
spectrum at the gear mesh frequency. It was the hypothesis
that gear system dynamics added phase error to the TSA for
the "local" analysis, which degraded the TSA and resulting in
an attenuation (due to filtering) of the gear mesh. This may
adversely affect the resulting gear analysis. Using a novel,
Figure 8 Comparison of the spectrum of the main idealized bandpass filter, a local tachometer signal was
gearbox using vibration for the tachometer derived for the shaft/gear under analysis, which significantly
Calculation of the Transfer Function due to Gear improved the TSA, improving the signal to noise ratio. In
Dynamics real-world examples from a light single helicopter, a 4 to 8 dB
improvement was observed in the TSA spectrum.
The transfer function, which represents the gear system
dynamics, can be calculated by using the input signal (e.g., Improving the signal to noise of the TSA will improve fault
the measured tachometer signal) and the "local" tachometer detection for HUMS applications. Further, the ability to do
signal. Consider that some transfer function H exists, then the analysis without a tachometer can help reduce the cost of
local tachometer response is the convolution in time of H with HUMS hardware. While a tachometer 1/Rev striker is needed
the input tachometer signal: for main and tail rotor balance, in general, no phase
information is required for drivetrain analysis. Besides the
𝑧𝑙𝑜𝑐𝑎𝑙(𝑡) = ℎ(𝑡) ∗ 𝑧(𝑡) eq. 9 improved analysis, the allows in the main gearbox, removing
a tachometer can reduce the HUMS kit cost, installation cost,
Taking in Fourier transform and maintenance cost on HUMS itself.

6
Author contact:

Eric Bechhoefer [email protected]

Yalin Ozturk [email protected]

REFERENCES
1. Bechhoefer, E., Kingsley, M, “A Review of Time
Synchronous Average Algorithms”, PHM Society
Annual Forum, 2009.
2. McInerny, S.A., Sun, Q., Hardman, B., "Public Domain
Algorithms for Bearing and Gear Fault Detection,"
Proceedings of the 11th International Conference on
Structure Borne Sound and Vibration, 2004.
3. Stewart, R.M., “Some Useful Data Analysis Techniques
for Gearbox Diagnostics,” Institute of Sound and
Vibration Research, Proceeding of Meeting on
Application of Time Series Analysis, Sept. 1977
4. Bechhoefer, E., He, D., "Improving Gear Fault Detection
by Reducing Tachometer Jitter," AHS Forum 72, May 1,
2014.
5. Bechhoefer, E., Van Hecke, B., He, D., Processing for
Improved Spectral Analysis, PHM Society Annual
Forum, 2013.
6. Ozturk, Y, “A Parametric Study on Planetary Gear
Dynamics” Thesis, The Graduate School of Natural and
Applied Sciences of Middle East Technical University,
2018.
7. Bonnardot, F., El Badaoui, M., Randal, R. B., Daniere,
J., Guillet, J., “Use of the acceleration signal of a gearbox
in order to perform angular resampling (with limited
speed fluctuations)," Mechanical Systems and Signal
Processing 19 (2005) 766-785.
8. Combet, F., Gelman, L., “An automated methodology for
performing time synchronous averaging of a gearbox
signal without speed sensor," Mechanical Systems and
Signal Processing 21 (2007) 2590-2606.
9. Braun, S., “The Extraction of Periodic Waveforms by
Time Domain Averaging”, ACUTSTIC, Vol. 32, 1975