You are on page 1of 11

International Journal of Fatigue 133 (2020) 105432

Contents lists available at ScienceDirect

International Journal of Fatigue


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

A rotating gear test methodology for evaluation of high-cycle tooth bending T


fatigue lives under fully reversed and fully released loading conditions

Isaac J. Honga, , Ahmet Kahramana, Neil Andersonb
a
Department of Mechanical and Aerospace Engineering, The Ohio State University, Columbus, OH 43210, United States
b
Mech. Systems, Externals and Nacelles, Pratt & Whitney Engines, East Hartford, CT 06108, United States

A R T I C LE I N FO A B S T R A C T

Keywords: Gear bending fatigue life is a primary consideration in gear design. Each gear tooth is subjected to cyclic root
Gears stresses that may cause fatigue fracture. A gear in a transmission is subjected to either fully released or fully
Stress measurement reversed loading conditions, with different fatigue lives. This study develops an experimental methodology for
Specimen design evaluation of high-cycle tooth bending fatigue lives of gears under both loading conditions. Unique specimens
High cycle fatigue
for tooth bending fatigue and an adaptive diagnostics method for test termination are designed while root strains
Fatigue test methods
under high-speed conditions are quantified. Example fatigue tests are presented to demonstrate the effectiveness
of the proposed methodology.

1. Introduction contact fatigue lives. This requires sophisticated analysis and testing
methods.
Long-term reliability of power transmission systems is critical in Gears in power transmissions operate in one of two modes. If the
many automotive and aerospace systems. As the primary power trans- gear serves as the input or output of the system (or it is the stationary or
mission components in these systems, gears must possess the required fixed member in case of a planetary gear set), then the gear acts as a
fatigue lives, often in excess of 100 million loading cycles. There are torque-splitting device where it drives or is driven by one or more gears
two types of material failure modes in gears. One is contact fatigue the same way. In this case, only one flank of a tooth on this gear is
(micro-pitting, macro-pitting or spalling), leading to degradation of loaded, resulting in fully released root fillet stress time histories. Central
contacting tooth surfaces. The tribology of the lubricated contact in- gears of a planetary gear set (sun and ring gears) are subject to such
terface and various multi-axial fatigue models have been combined to loading. The second type of loading occurs when the gear is an idler
predict macro and micro-pitting of gear and representative roller sur- (i.e. with no input, output or fixed member assignments), simply
faces [1–4]. On the experimental side, standardized test set-ups and transmitting power from one gear to the other (e.g. planet gears of a
associated methodologies [5–7] have been widely used to evaluate planetary gear set). In this case, the gear teeth come to contact with the
contact fatigue lives of gears under actual rotating conditions to gen- mating gears along opposite flanks, subjecting tooth root fillets to fully
erate long-cycle contact fatigue data. reversed stress time histories. Conventional gear design methods often
The second gear material failure mode is tooth bending fatigue. A do not distinguish these different loading conditions, or employ dif-
gear tooth has an approximate shape of a cantilevered tapered plate ferent allowable stress levels based on field experience. Using a pla-
with a root fillet geometry that creates a stress concentration. Each time netary arrangement as an example, it can be hypothesized that the fully
a tooth of a gear passes through mesh, it experiences a sudden tooth reversed loading conditions of planets are more severe than the fully
force for a brief period, followed by a much longer period of zero load released loading on the sun and ring gears due to the added compres-
before it returns to the gear mesh. Such cyclic loading results in fatigue sive stress cycle. As such, the planet gears might be expected to require
of the tooth such that cracks initiated along root fillet lead to tooth a bending fatigue life penalty if they are designed to the same stress
fracture. While contact fatigue failures occur gradually and can be de- levels as the sun and ring gears. It becomes important to be able to
tected before the power transmission system fails, tooth fractures are quantify this penalty experimentally and apply this information to
catastrophic, resulting in failure of the whole system. In order to avoid predict fatigue lives under both types of loading conditions.
this risk, the designer must develop gears with required bending and While there are many published experimental works in gear


Corresponding author.
E-mail address: hong.250@osu.edu (I.J. Hong).

https://doi.org/10.1016/j.ijfatigue.2019.105432
Received 15 September 2019; Received in revised form 8 November 2019; Accepted 19 December 2019
Available online 20 December 2019
0142-1123/ © 2019 Elsevier Ltd. All rights reserved.
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

Nomenclature T Torque transmitted by the outside loop of the test machine


Zg Number of teeth on the outside mating gears
Symbol Definition Zi Number of teeth on the central test pinion
a(i) RMS vibration level at the state i α Empirical slope parameter for defining the adaptive ac-
(i)
astop Adaptive acceleration limit celeration limit
 (Ωg )
A Frequency response curve calculated RSS of mesh har- Δa Empirical offset parameter for defining the adaptive ac-
monic amplitudes celeration limit
Aj Amplitude of the j-th gear mesh harmonic εmax Maximum measure strain at dynamic conditions
fni Natural frequencies of the test machine εmax, s Maximum measured strain at quasi-static conditions
fm Mesh frequency εmin Minimum measured strain at dynamic conditions
RSS Root sum squared calculation εmin, s Minimum measured strain at quasi-static conditions
i Index of discrete time step for adaptive acceleration limit R Ratio of the minimum to maximum root bending stress in
calculation a stress cycle
j Integer index of mesh frequency harmonics σ1,max Maximum normalized 1st principal stress
Kvt (Ωg ) Dynamic loading factor for tensile strains σeq, cr Normalized equivalent Goodman completely reversed
Kvc (Ωg ) Dynamic loading factor for compressive strains stress amplitude
L Baseline fatigue test torque σmax Maximum root normal bending stress
RL Root fillet radius σmax Minimum root normal bending stress
Rr Root circle radius
t Time

bending fatigue, they are incomplete. In assessing the relevance of a [9] used a full planetary set to experimentally investigate the bending
gear tooth bending study, one must consider the following. (i) Were fatigue lives in a helical planet gear. The speed of the gear set in these
actual gears used in the experiments? (ii) If gears were used, were they experiments was low, and hence, duration of these tests was limited to
operated under normal rotating conditions in mesh with a mating gear one million cycles. Hatano et al. [20] used a power-circulation type
(or gears)? (iii) In rotating gear experiments, were the effects of dy- gear fatigue tester to investigate the effect of shot peening on the
namic behavior on the root stresses due to the vibratory conditions of bending fatigue life of gear teeth. However, no details of the test ma-
the test set-up included? chine were given and very few details of the test gears were provided.
Early experimental fatigue studies used bar specimens loaded in 4- Tests were suspended at 10 million cycles and no information is pro-
point bending as a substitute for a gear tooth root [8–10]. These tests vided on how root stress values were obtained. One main reason for the
involved machining a notch into the bar in an attempt to simulate the lack of rotating gear bending fatigue studies is that they require a re-
tooth root stresses. They were capable of creating a sinusoidal reversed liable diagnostic method to stop the test before catastrophic failure of
stress load ratio of R = σmin σmax = −1 at modest frequencies of around the test gear and test set-up occurs. Also challenging is to design a test
10 Hz. These studies were useful in terms of evaluating different gear gear that has tooth bending as its primary failure mode, avoiding other
steel types under cyclic loading, but did not represent gear loading typical failures such as surface wear, contact fatigue (pitting) or
conditions well. Another group of studies [11–17] used variations of a scuffing. Accordingly, the objectives of this study are as follows:
single-tooth bending (STB) test [18] in an attempt to bring the loading
conditions closer to that of a gear mesh by using an actual gear. In these • Design and develop a new test set-up to investigate the bending
tests, a spur gear is placed on a stationary fixture that reacts one tooth fatigue lives of spur gear teeth under both R = 0 and R = −1 type
while applying a fully released cyclic loading (R ≈ 0 ) to another tooth loading conditions. This set-up must be reconfigurable to apply both
to induce fatigue failures. Two main shortcomings of this method are types of loading to the same gear specimen design in order to collect
that (i) it is difficult to produce reversed loading on a gear tooth since comparable long-cycle fatigue data (say, in excess of 100 million
the fixture can only push down (or up) on the tooth, and (ii) it cannot loading cycles).
apply the actual load cycle since the loading occurs at a fixed point (roll • Design test gear specimens to operate under both loading conditions
angle) [17]. Furthermore, most of these STB tests employ a servo-hy- with tooth bending fatigue as the primary failure mode.
draulic load frame that operates at low loading frequencies, limiting the • Quantify the dynamic influences of the set-up on gear stresses such
run-out life to 1–10 million cycles. This produces only the lower-cycle that they are included in evaluation of the fatigue data since the test
and high-stress portion of a stress-life curve while the real need in most set-up must operate at a high speed to allow accumulation of
automotive and aerospace applications is in excess of several hundreds loading cycles in accelerated rates.
of million cycles. • Develop an effective diagnostic system in order to capture the life of
A unique test setup utilizing two non-rotating gears meshed to- the first failed tooth of the test gear with no failure to the remaining
gether was developed by Kodeeswaran et al. [19] in order to study the teeth.
bending fatigue performance of plastic and composite gears subjected
to fully-reversed and fully-released loading. This setup utilized a servo This paper describes the implementation of these four experimental
motor to “shake” a steel driver gear against the fixed test specimen to objectives towards development of a robust methodology to evaluate
produce the cyclic tooth loading by contacting a single test gear spe- long-cycle bending fatigue lives of spur gear teeth under both types of
cimen tooth against either a single adjacent driver tooth or both ad- loading. A companion paper [21] will employ the methodology pro-
jacent driver teeth. Resultant loading is similar to those produced in a posed in this study to perform fatigue tests under R = 0 and R = −1
STB test as only one mesh position is simulated. While the setup is able type loading conditions to (i) determine differences in the measured
to produce fully-reversed loading, it is limited by the capabilities of stress-life curves of different loading conditions empirically, and (ii)
servo motors to produce larger torques required to test steel gears and provide means to assess the accuracy of various multi-axial fatigue
was limited to 5 Hz loading frequency in the fully-reversed condition. criteria when applied to the gear tooth bending fatigue problem. The
Gear literature is extremely sparse in terms of rotating gear tooth goal of this paper is to develop a methodology to allow evaluation of
bending experiments. As one of a few studies of this kind, Phelan et al. the impact of properties regarding metallurgical and mechanical

2
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

characteristics of the specimens on tooth fatigue performance. These support rolling element bearings were designed to have significantly
properties include not only the material types but also other effects higher lives that the test pinion. The gears of the reaction gearbox were
associated with heat treatment, material cleanliness, surface lay and designed to contact and bending stresses that are significantly lower
roughness, root surface finishing processes (hobbed, ground or super- than those of the test gears.
finished) as well as processes inducing residual stresses (shot peening,
etc.). The proposed methodology intends to quantify such effects, pro- 2.2. Design of the test gears
vided sets of specimens can be procured to represent these variations.
Since the main goal in this study was to compare bending fatigue
2 Development of an experimental methodology lives of a gear under the R = 0 and R = −1 loading conditions, the
weakest gear must be the pinion meshing with two identical side gears.
2.1. Test machines The task here was to design test gears such that the tooth breakage was
the primary failure mode while gear meshes maintained their integrity
Gear contact fatigue test machines [5–7] form the basis for the (i.e. did not exhibit pits or excessive wear) along their contact surfaces.
design of a new set-up suitable for testing of both fully released and There are two primary ways to increase tooth root stresses while
fully reversed loading conditions. Test and reaction gearboxes, have a keeping contact stresses the same. One way is to reduce the fillet radius
pair of spur gears of the same center distance and gear ratio that are in the root to increase the stress concentration and hence the maximum
connected to each other via flexible shafts to form a closed power cir- root stress. A conventional gear pair was first designed with a con-
culation loop. In this back-to-back arrangement, one of the connecting ventional root radius Rr representing typical dedendum height as
shafts is split by a coupling that acts as the loading device. A certain shown in Fig. 6. In this design, a full circular root fillet with the largest
constant torque T is applied to the loop mechanically through this split possible root fillet radius RL = Rmax resulted in a very modest value of
coupling, which is the torque transmitted by both gear pairs. As such, a maximum root stress. While it is possible to increase the maximum root
small drive motor connected to one of the shafts of the reaction gearbox fillet bending stress σmax by reducing RL , an extremely small value was
has the sole duty of overcoming efficiency losses to rotate the set-up at a required to achieve desired σmax values. As this represents a severe
desired speed. stress concentration and a very steep gradient of root stresses along the
The new machine set-up in the R = −1 loading configuration, as root fillet, any minor deviations from this root radius in manufacturing
shown in Fig. 1(a), consists of three parallel shaft assemblies with the the pinion would result in sizable differences In σmax , adding sig-
two outer shafts used to form the power circulation loop while one shaft nificantly to the scatter of the resultant fatigue lives.
contains the split coupling loading device. The middle shafts of the test The second way to increase root stresses while maintaining
and reaction gearbox are not connected to each other. As such, the equivalent contact stresses is to reduce Rr (i.e. increase dedendum
middle gear (called pinion here) meshing with the two side gears is an height) without resorting to the option of reducing RL . Root fillet
idler gear. Each idler gear has opposite flanks of its teeth in contact with bending stress is inversely proportional to Rr . Lower values of Rr allow
the two side gears as it acts as the driver for one of its meshes and the lower applied torques which in turn reduce stresses on other machine
driven gear for the other. Gear mesh forces are the same for both me- components. Table 1 specifies base geometric parameters of the test
shes of a gearbox, and as such, any given point along the root of the gear pairs (pinon and a side gear with both side gears being identical in
idler gear experiences a tensile stress when the tooth is at one mesh
while a compressive stress of a similar magnitude is produced when the
same tooth is passing through the other mesh. The resultant, cyclic
stress time history for this configuration is shown schematically in
Fig. 1(b). The R = 0 loading arrangement shown in Fig. 2(a) is obtained
by moving one of the outer connecting shafts to the middle axis to
connect the middle gears. There is only one gear mesh per gearbox in
this configuration, resulting in a single tensile peak on the tensile side of
the root fillet per each gear revolution as illustrated in Fig. 2(b).
Fig. 3(a,b) shows the top view of the test machine in the R = −1
and R = 0 loading configurations. A solid model isometric view of the
final machine set-up is shown in Fig. 4(a) in the R = −1 configuration.
Since long-cycle fatigue is a primary objective, tests are expected to run
for very long periods even when the machines are operated at elevated
speeds. As such, three of these machines were fabricated in this study to
increase throughput as shown in Fig. 4(b). Each machine is driven by an
AC motor connected to the machine input shaft directly through a
double flex-disc type coupling. Fig. 5 illustrates the method of applying
a constant torque to the power circulation loop in the R = −1 loading
configuration.
Within the speed ranges of operation, dip lubrication was deemed
appropriate and effective, also considering that the main impact of the
lubrication system is on contact fatigue, not on the bending fatigue. An
ISO 150 gear oil was chosen as the lubricant. Static oil level was set to
be just below the shaft axes to ensure adequate entrainment of lubricant
into the mesh while limiting amount of oil churning. Sump tempera-
tures were maintained at within 90 ± 3°C using finned copper heat
exchangers placed along the bottom of the gearbox with circulating
chilled water running through them.
As the goal here was to have the tooth breakage of one of the middle Fig. 1. Fully reversed (R = −1) loading configuration with (a) its power cir-
gear (pinion) of the test gearbox as the primary failure mode, all other culation schematic and (b) a schematic of the resultant root stress time history
components the machine including the connecting shafts, couplings, at a point on the root fillet.

3
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

Fig. 2. Fully released (R = 0 ) loading configuration with (a) its power circu-
lation schematic and (b) a schematic of the resultant root stress time history at a
point on the root fillet.

Fig. 4. (a) Isometic view of the the test set-up in the R = −1 loading config-
uration with the safety covers removed, and (b) gear tooth bending test facility
consisting of three test set-ups.

Fig. 3. Top-view of test machine in (a) fully reversed (R = −1) and (c) fully
released (R = 0 ) configurations. Safety covers and gearbox covers are removed
for demonstration purposes.

Fig. 5. Method of applying a constant torque to the power circulation loop in


the R = −1 loading configuration.

4
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

Fig. 6. Methods of increasing root stresses through either reducing the root
fillet radius RL or reducing root circle radius Rr .

Table 1 Fig. 8. Deformable-body finite element models in (a) fully reversed (R = −1)
Parameters of a test gear pair formed by a middle pinion ans am mationg side and (b) fully released (R = 0) configurations.
gear. All dimensions in mm unless specified.
Parameter Test pinion Mating gear methods exist ranging from standards-based simplified predictions to
semi-analytical load distribution models or full-scale deformable-body
Number of teeth 17 25 gear contact models. Here, a commercial deformable-body gear contact
Module (mm) 4.3535
model (Transmission3D [22]) was employed to predict the state of root
Pressure angle (Degrees) 26.842
Center distance (mm) 91.5
stresses by considering all gears in mesh as deformable bodies. Fig. 8
shows deformable-body contact models of both loading configurations
that illustrates the type of loading conditions produced in the test
machine. This model has a sophisticated contact solver, specifically
designed to predict the load distributions and stresses of gears of entire
transmissions. The model employs a hybrid approach by matching a
semi-analytical formulation at the contact zone with finite quasi-prism
elements employing 10th order Chebyshev polynomial expansions at
each intermediate axode within each element. Refs [23,24] provide the
theoretical details of this model.

2.3. Vibration-based diagnostics of failures

Ideally, a test of this kind would be stopped when a single tooth has
either a sizable crack, as illustrated in Fig. 9(a), or a single tooth breaks
off. Considering that these tests involve very long time spans before
failure, tests must be performed at high rotational speeds. Even at these
Fig. 7. A set of gears in the R = −1 configuration with side gearbox cover re- high speeds, a fatigue test of this kind can take a few months of con-
moved. tinuous operation. As such, it is not feasible to monitor and suspend a
test manually when a pinion reaches the conditions shown in Fig. 9(a).
the R = −1 loading configuration). Fig. 7 shows an example set of three A gear tooth loses a sizeable portion of its load carrying capacity once a
gears mounted in the test gearbox in the R = −1 configuration. crack along the root of a tooth reaches a certain size. This is true even
An accurate prediction of the state of stresses along the root fillets is when the crack is still small relative to the cross-sectional area of the
critical to interpretation of fatigue test results. Various computational tooth along its failure plane. Since the tooth with the crack continues to
run, the crack propagates very rapidly causing a complete fracture of

5
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

of the accelerometers during the first two-hour long segment of testing.


Even after the significant changes within the first 20 min of the test
attributable to stabilization of the gearbox temperature, the RMS vi-
bration level is observed to vary considerably instead of settling to a
constant average level. Other factors not related to temperature, such as
gradual wear of contacting surfaces, contribute to changes in vibration
levels as well. As such, using a constant (static) baseline RMS level from
which a shutdown threshold is defined either resulted in false positives
or failed to stop the test at the required condition. In order to resolve
this issue, a dynamic, adaptive vibration limit was defined as a function
of the running average of the RMS vibration level to trigger the machine
to shut down if exceeded. This adaptive vibration limit was designed to
ignore the slow changes (drift) in RMS vibration levels throughout the
test but capture any sudden change or spike in vibration. These spikes
occur as a crack rapidly propagates through the tooth root or when the
tooth breaks off. The equation governing this approach is given as

(i) (i − 1)
astop = α [astop − Δa] + (1 − α ) a(i) + Δa (1)
(i)
where astop is the value of the acceleration limit in effect during the
current state i of the test, a(i) is the RMS vibration level at the current
(i − 1)
state i, and astop is the value of the acceleration limit at the previous
state i − 1. Here, the incremental increase Δa from the baseline an-
ticipated due to the onset of the failure and parameter α ∈ [0, 1] are
defined empirically. By setting the value of α close to 1.0, this equation
weights the previous value of the limit heavily while allowing it to
change by a small proportion of the current RMS vibration value. The
value of α is also dependent on the update rate of the current state i, the
time history of Fig. 11 shows baseline vibration level a(i) and the
(i)
adaptive limit astop . A sudden increase in vibration level due to the onset
Fig. 9. Example of pinion at the end of a test: (a) with a single tooth of a pinion
exhibiting a sizable crack and (b) a pinion with all of its teeth broken.

the tooth. If the machine fails to stop when the pinion reaches this state,
the loss of one tooth cascades the failure to all other teeth on the pinion
due to increased loading caused by missing teeth. This leads to total
destruction of the pinion and its mating gears, and imposes undesirable
loading on test gear shafts and bearings. One example of such cata-
strophic failure is shown in Fig. 9(b) where all of the teeth of the pinion
are broken. On this failed pinion, it is almost impossible to determine
which tooth failed first and where the crack was initiated since broken
teeth disturb fracture sites. As such, it is vital that an automated diag-
nostic method be developed to (i) detect when a failure state as shown
in Fig. 9(a) is reached, and (ii) stop the machine in a reasonable amount
of time such that the failure does not progress to the state shown in
Fig. 9(b).
In order to detect a failure at the state shown in Fig. 9(a), piezo-
electric accelerometers were mounted on the outside of the gearbox in
various locations. After extensive trials, it was found that two of these
locations were effective in providing robust levels of vibrational re-
sponse, one near the support bearings and one at the center of the top
cover of the gearbox. They are both shown in Fig. 10. These accel-
erometers were both PCB 353B18 accelerometers with a measurement
range of ± 500 g and a frequency range of 1–10,000 Hz on ± 5%.
Signals from the accelerometers were fed into a signal conditioner,
which was set to output an RMS vibration level.
One simple way to use these vibration signals for diagnostics is to
set an upper limit for the measured root-mean-square (RMS) vibration
amplitudes that represent a failure state as the ones shown in Fig. 9(a).
The machine would then be commanded to stop immediately once this
threshold is reached. This limit would be set as a certain percentage
increase over the baseline RMS levels at the beginning of a test. This
simple method was observed to be ineffective as the baseline RMS le-
vels measured during any given fatigue test were observed to evolve.
Fig. 11 shows an example of a measured RMS vibration level from one Fig. 10. Accelerometers mounted on the test gearbox for diagnosing the failure
(a) on the top cover plate and (b) on the side bearing cover.

6
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

(i)
Fig. 11. An example of measured acceleration signal a(i) and the dynamic acceleration threshold astop established using Eq. (2.1) with α = 0.98 and Δa = 6g . The
(i)
sudden spike of a(i) at the end results in the condition a(i) ⩾ astop for the test to be terminated.

(i)
of tooth breakage resulted in the condition where a(i) ⩾ astop to trigger correspond to the fundamental gear mesh frequency
(i) fm = Zg Ωg 60 = Zi Ωp 60 and its higher harmonics jfm where Zg = 25
the machine to shut down. Here, astop established using parameter va-
lues of α = 0.98 and Δa = 6g were effective in capturing failures with teeth, Zi = 17 teeth and j is an integer ( j > 1). Such diagonal gear mesh
only the first tooth cracked or broken off. order lines, at least up to j = 10 , are clearly evident, suggesting that
they contribute to the response. Also noted along the horizontal fre-
quency axis are a number of vertical bands where the vibration levels
2.4. Quantifying dynamic effects increase, particularly at fni = 2, 800 and 4,800 Hz, representing a subset
of natural frequencies of the machine that are excited by the gear
The Finite Elements based contact models presented in Fig. 8 predict
tooth-bending stresses under quasi-static conditions without dynamic
effects. If dynamic effects exist at the operating conditions used to
perform fatigue tests, these quasi-static stresses will be different from
those under dynamic conditions. The test machines shown in Fig. 4(b)
are dynamic systems with their own natural modes and resultant vi-
bration resonance peaks. Operation of the machines at higher speeds
might increase gear mesh and tooth forces beyond what is predicted or
measured at very low speed conditions. These increases could even be
more significant if the operating speed is such that gear mesh frequency
harmonics coincide with one of the natural frequencies of the test
machine [25–28]. As such, it is critical (i) to choose operating condi-
tions that do not result in significant dynamic effects by staying away
from significant resonance regions, and (ii) to quantify dynamic am-
plification of root stresses at chosen operating conditions away from the
resonance peaks. These dynamic amplification factors would then be
included in the stress values used in stress-life data or Weibull dis-
tributions.
The first task of identifying the resonance frequencies was done
conveniently by using the accelerometers shown in Fig. 10 as part of the
diagnostics system. The accelerometer signals were processed at a high
sampling rate to perform transient and steady-state characterization of
the test machines. The machine was ramped slowly at a 40 rpm/s sweep
rate on the pinion from a standstill to a maximum speed of Ωg = 3, 500
rpm (corresponding to a pinion speed of Ωp = 5, 147 rpm) and then
slowly decelerated at the same rate to a stop again. In contrast to
dedicated gear dynamics studies [25–27] where more direct measure-
ments such as dynamic transmission error were employed, the mea-
surements here relied on housing vibrations. As such, they were indirect
in terms of assessing the vibration source strength, but still sufficient to
assess the modal characteristics of the machine. An example waterfall
plot is shown in Fig. 12 for a machine in the R = −1 loading config-
uration of Figs. 1 and 3(a) under the lowest test torque level L. In this Fig. 12. Waterfall plots of acceleration data recorded from the gearbox housing
figure, several diagonal lines with higher vibration amplitudes for (a) sweep-up and (b) sweep-down conditions.

7
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

meshes. Resonance conditions are evident when jfm ≈ fni . Most severe
of these conditions is when 2fm ≈ fn1 = 2, 800 Hz. There are two no-
ticeable horizontal bands of high vibration with a sharp lower boundary
at Ωg =3,260 and 1,640 rpm in the sweep-up plot and at Ωg = 3, 150 and
1,600 rpm in the sweep-down plot, indicating that tooth separations
may take place in these resonant speed values.
Each frequency spectrum at each Ωg increment in Fig. 12 can be
represented by a single acceleration amplitude that is the root-sum-
square (RSS) of the mesh harmonic amplitudes, defined here by using
the first five mesh harmonics as

5
 (Ωg ) =
A ∑ Aj2
j=1 (2)

where Aj is the amplitude of the j-th gear mesh harmonic. As the spectra
shown in Fig. 12 do not consist only of pure gear mesh tones, but also
some sideband activity, Aj was taken as the maximum amplitude within
a narrow band [0.85jfm , 1.15jfm ].
Fig. 13 shows the resultant A  (Ωg ) curve corresponding to the data
of Fig. 12 with both sweep-up and sweep-down data points overlaid.
Here, three significant resonance peaks are evident at Ωg =3,360, 2,240
and 1,680 rpm, corresponding to the gear mesh frequencies of
fm = 1, 400, 933 and 700 Hz. This indicates that all three resonance
peaks occur as harmonic orders 2fm , 3fm and 4fm excite the same mode at
fn1 = 2, 800 Hz (i.e. when 2fm ≈ fn1, 3fm ≈ fn1 and 4fm ≈ fn1, respec-
tively). A softening type nonlinearity is evident for the resonance peaks
at Ωg =3,360 and 1,680 rpm, corresponding to broadband spectra
forming the horizontal bands in Fig. 12. As the goal was to operate the
test machines at maximum possible off-resonance condition, a speed of
Ωg =3,000 rpm was deemed reasonable for the fatigue tests, as it was the
highest speed that was safely away from the largest resonance peak of
the machine. Similar vibration responses were measured at other load
levels as well as for the fully released loading condition.
Root strain measurements were completed next to quantify dynamic
effects on root stresses directly. For this, root stress measured at quasi-
static (very low) speeds must be compared to those measured at oper-
ating test speeds to quantify dynamic factors. As shown in Fig. 14(a),
strain gages were placed in the root of the pinion in order to measure Fig. 14. (a) Strain gages installed on a test pinion and mounted on the pinion
bending strain under operation. These gages were Micro-Measurements shaft, and (b) slip ring vibration isolation fixture devised to transfer strain
Brand P/N EA-06-031EC-350 and professionally mounted by experi- signals to the fixed frame.
enced technicians at specific known location in the gear tooth root and
oriented in the direction of bending. Fig. 14(b) shows the set-up devised
to transfer the measured strain signals from the rotating pinion to the

 (Ωg ) .
Fig. 13. Measured forced response A

8
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

fixed frame. It consisted of a slip ring mounted at the end of a short


shaft that was supported by a rolling element bearing. Some earlier
studies (e.g. ref. [26,28]) mounted the slip ring directly to the end of
the pinion shaft, which caused noise during transfer of the data through
slip ring due to vibration of the rotating shaft. In the set-up of
Fig. 14(b), the slip ring shaft is connected to the pinion shaft via a
flexible tube, thus isolating the slip ring from the gearbox vibrations
completely.
Strain measurements were first taken at a very low speed of
Ωg =150 rpm to obtain the static baseline strain from each gage location
at that load level. Next, strain data was collected at the desired test
speed. Several steps were taken in the data analysis to reduce the
measured data to dynamic stress magnification factors. First, a finite
impulse response (FIR) low-pass filter was applied with a cutoff fre-
quency of 5 kHz. Secondly, the DC offset in the signal was removed by
subtracting the median value of the entire time history. For each data
segment at a particular speed, an algorithm was applied to locate the
maxima for tension and minima for compression, of strain for each
rotation of the pinion. Fig. 15(a) shows a segment of filtered strain data
with maximum and minimum values for each rotation. A comparison of
a pair of dynamic (Ωg =3,000 rpm) and quasi-static (Ωg =150 rpm) strain
is shown in Fig. 15(b) for one complete revolution of the pinion. While
these signals repeat themselves quite well, slight variations in each
revolution is expected as the gaged tooth meshes each time with a
different tooth of the mating gear (at a ratio of 17:25, 425 pinion ro-
tations are required for the same pair of teeth to come to contact again).
Therefore, maxima and minima of each signal was examined statisti-
cally. The corresponding histograms showed that the distributions of
maximum and minimum strains achieved in each mesh cycle of a tooth
are unimodal and symmetric such that the mean, median and mode
values are all approximately equal. The mean of the each distribution
was used to define a dynamic factor that represents the change in
Fig. 16. Kvt (Ωg ) curves of a pinion under fully reversed conditions at two dif-
maximum or minimum strain achieved in the gear tooth root due to
ferent load levels of (a) L and (b) 1.14L.
dynamic loading effects. Defining the mean maximum and minimum
strain values at a given speed Ωg as εmax and εmin, and the corresponding
values at quasi-static conditions (Ωg =150 rpm) as εmax, s and εmin, s, dy- 0.987 depending on load for the fully released loading at the same
namic factors for the pinion tensile and compressive strains are given as speed.
In addition to evaluating the steady-state dynamic factors at a given
εmax εmin speed, root strains were also measured under transient conditions as the
Kvt (Ωg ) = , Kvc (Ωg ) = .
εmax, s εmin, s (3a,b) machine speed was increased slowly. Fig. 16 shows Kvt (Ωg ) curves ex-
tracted from these slow, speed sweep-up tests at two load levels: L and
The dynamic factors calculated from the measurements at
1.14L. Also shown on these figures are the Kvt values at Ωg = 3, 000 rpm
Ωg =3,000 rpm at various torque values for both loading configurations
that were measured earlier under steady-state conditions. When viewed
will be used in our companion paper [21] to incorporate the dynamic
together, Figs. 13 and 16 confirms that selection of Ωg = 3, 000 rpm as
factors in fatigue tests. For the fully reversed loading, dynamic factors
the selected test speed was appropriate as it is away from any re-
at Ωg =3,000 rpm ranged from 1.02 to 1.21 indicating that the operation
sonances regardless of the torque transmitted. The fatigue tests pre-
at this speed results in dynamic stresses that are 2–21% higher than
sented in Ref. [21] were performed at this speed and the dynamic
their static counterparts. Meanwhile, Kv ranged between 0.94 and

Fig. 15. (a) A filtered strain signal for the fully reversed loading with the local maxima and minima for each pinion revolution identified, and (b) a comparison of the
strain time histories measured by a gage under quasi-static (Ωg = 150 rpm) and dynamic (Ωg = 3, 000 rpm) conditions.

9
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

Fig. 17. Examples of fracture surfaces from two fully reversed tests showing a fish eye initiation site; (a) a test at 1.33L and a life of 11.8 M cycles, and (b) a test at
1.14L and a life of 94.5 M cycles.

Fig. 18. Examples of fracture surfaces from two fully released tests showing a fish eye initiation site; (a) a test at 1.69L and a life of 34.2 M cycles, and (b) a test at
1.85L and a life of 25.9 M cycles.

factors associated with this speed under different load levels and counterparts, creating a two sloped S-N curve with a reduction in the
loading configurations were factored in the resultant stress values. fatigue limit at which infinite life is assumed. Initiation sites in
Fig. 17(a) and (b) were 0.105 and 0.040 mm below the surface, re-
3. Example fatigue tests spectively, while they were 0.145 and 0.070 mm below the surface for
Fig. 18(a) and (b).
In a companion paper [21], the experimental methodology pre- In addition to digital inspections of the fracture surface, CMM
sented here was employed to develop extensive data sets for both measurements along the root fillets of broken teeth were performed to
R = −1 and R = 0 loading conditions. Statistical analyses of these re- quantify the radii at which the teeth broke. These measurements are of
sults were used to determine a stress factor between the two loading interest in correlating to any predictions of the location of crack in-
conditions for the same number of cycles to failure. A number of multi- itiation from a fatigue life model. Tests shown in Fig. 17(a) and (b) had
axial fatigue life models with basic material fatigue data and residual their cracks initiate at radii of 27.97 and 27.30 mm while the tests of
stress measurements were used to simulate these experiments to assess Fig. 18 cracked at radii of 25.83 and 27.81 mm, all measured from the
accuracy of such predictions. Four examples of these experiments will center of the pinion to the root fillet location where crack formed.
be shown here as examples, two at each load configuration.
Digital inspections of fracture surfaces from two R = −1 tests are 4. Conclusion
shown in Fig. 17. The test in Fig. 17(a) was at a higher load level of
1.33L and lasted 11.8 M cycles while the test in Fig. 17(b) at 1.14L This study was concerned with development of a new experimental
lasted 94.5 million cycles. Likewise, fracture surfaces from two R = 0 methodology to evaluate tooth bending fatigue lives of gears under two
tests are shown in Fig. 18(a) for a gear specimen failed at 1.69L after different loading conditions. As the focus was on long-cycle fatigue
34.2 M cycles and in Fig. 18(b) for another specimen failed at 1.85L performance under modest gear mesh force conditions, a rotating gear
after 25.9 M cycles. All four of these contained “fisheye” initiation sites set-up was designed and developed that allowed a test gear to assume
[29–31], which were typically at subsurface locations as opposed to idler (fully reversed) or torque-split (fully released) duties. A test gear
surface originated failures. Several investigations into very high cycle specimen was designed to ensure that its most likely failure mode is
fatigue [30–34] showed that the fatigue lives associated with subsur- tooth fracture. Accelerometer and strain gage based measurement sys-
face failures are indeed different than the surface initiation tems were employed to determine optimal operating speed conditions

10
I.J. Hong, et al. International Journal of Fatigue 133 (2020) 105432

as well as the dynamic factors at the selected test condition. The same helicopter case carburized gears: influence of material, design and manufacturing
vibration measurement system was also facilitated to devise a diag- parameters. Gear Technol 2009:68–76.
[13] Handschuh RF, Krantz T, Lerch BA, Burke CS. Investigation of low-cycle bending
nostic system that detects tooth failures long before total destruction of fatigue of AISI 9310 steel spur gears. NASA TM 214914; 2007. doi:10.1115/
the test specimens. Examples of fully reversed and fully released fatigue DETC2007-34095.
tests were presented at the end to demonstrate the effectiveness of the [14] Medlin D, Krauss G, Matlock DK, Burris K, Slane M. Comparison of single gear tooth
and cantilever beam bending fatigue testing of carburized steel. SAE Pap 950212;
proposed methodology. 1995. doi:10.4271/950212.
A companion paper by the same authors [21] employs this metho- [15] Savaria V, Bridier F, Bocher P. Predicting the effects of material properties gradient
dology to conduct families of fully reversed and fully released long- and residual stresses on the bending fatigue strength of induction hardened aero-
nautical gears. Int J Fatigue 2016;85:70–84. https://doi.org/10.1016/j.ijfatigue.
cycle rotating gear tests. These sets of data are analyzed statistically to 2015.12.004.
form stress-life curves for each loading condition to define a stress [16] Shen T, Krantz T, Sebastian J. Advanced gear alloys for ultra high strength appli-
factor that would correlate relative root stresses of a gear under both cations. NASA TM 217121; 2011.
[17] Sanders A, Houser DR, Kahraman A, Harianto J, Shon S. An experimental in-
conditions to yield the same fatigue life. Furthermore, this companion
vestigation of the effect of tooth asymmetry and tooth root shape on root stresses
study employs a deformable-body gear stress prediction model, a set of and single tooth bending faituge life of gear teeth. ASME Int. Power Transm.
basic material data and a number of multi-axial fatigue criteria to si- Gearing Conf., Washington D.C.. 2011.
mulate fatigue experiments in order to assess their accuracy. [18] SAE international surface vehicle recommended practice. Single tooth gear bending
fatigue test. SAE Stand J1619; 1997.
[19] Kodeeswaran M, Verma A, Suresh R, Senthilvelan S. Bi-directional and uni-direc-
Declaration of Competing Interest tional bending fatigue performance of unreinforced and carbon fiber reinforced
polyamide 66 spur gears. Int J Precis Eng Manuf 2016;17:1025–33. https://doi.org/
10.1007/s12541-016-0125-6.
The authors declare that they have no known competing financial [20] Hatano A, Namiki K. Application of hard shot peening to automotive transmission
interests or personal relationships that could have appeared to influ- gears. SAE Pap 938179; 1992. doi:10.4271/920760.
ence the work reported in this paper. [21] Hong Isaac J, Kahraman A, Anderson N. An experimental and theoretical study of
long-cycle bending fatigue lives of gears under fully released and fully reversed
loading conditions, in review. Int J Fatigue 2019.
References [22] Transmission 3D. Hilliard, OH: Advanced Numerical Solutions LLC; 2007.
[23] Vijayakar S, Busby HR, Houser DR. Finite element analysis of quasi-prismatic bodies
using chebyshev polynomials. Int J Numer Methods Eng 1987;24:1461–77.
[1] Li S, Kahraman A. A fatigue model for contacts under mixed elastohydrodynamic
[24] Vijayakar S. A combined surface integral and finite element solution for a three-
lubrication condition. Int J Fatigue 2011;33:427–36. https://doi.org/10.1016/j.
dimensonal contact problem. Int J Numer Methods Eng 1991;31:525–45. https://
ijfatigue.2010.09.021.
doi.org/10.1002/nme.1620310308.
[2] Li S, Kahraman A. A physics-based model to predict micro-pitting lives of lubricated
[25] Kahraman A, Blankenship GW. Experiments on nonlinear dynamic behavior of an
point contacts. Int J Fatigue 2013;47:205–15. https://doi.org/10.1016/j.ijfatigue.
oscillator with clearance and periodically time-varying parameters. J Appl Mech
2012.09.002.
1997;64:217–26.
[3] Li S, Kahraman A. Micro-pitting fatigue lives of lubricated point contacts: experi-
[26] Talbot D, Sun A, Kahraman A. Impact of tooth indexing errors on dynamic factors of
ments and model validation. Int J Fatigue 2013;48:9–18. https://doi.org/10.1016/
spur gears: experiments and model simulations. J Mech Des 2016;138. https://doi.
j.ijfatigue.2012.12.003.
org/10.1115/1.4034175. 093302–093302–13.
[4] Li S, Kahraman A. A micro-pitting model for spur gear contacts. Int J Fatigue
[27] Anichowski, Brian J. An experimental investigation of the effect of spacing errors on
2014;59:224–33. https://doi.org/10.1016/j.ijfatigue.2013.08.015.
the loaded transmission error of spur gear pairs. The Ohio State University; 2017.
[5] ISO. ISO 14635-1:2000 - Gears – FZG Test Procedures – Part 1: FZG Test Method A/
[28] Anichowski, Brian J, Talbot D, Kahraman A. Dynamic transmission error mea-
8, 3/90 for relative scuffing load-carrying capacity of oils. Geneva: International
surements from spur gear pairs having tooth indexing erros. ASME Power Transm.
Organization for Standardization; 2000.
Gearing Conf. DETC2017-67314, Cleveland; 2017.
[6] ISO. ISO 14635-2:2004 - Gears – FZG Test Procedures – Part 2: FZG Step Load Test
[29] Stanzl-Tschegg S. Very high cycle fatigue measuring techniques. Int J Fatigue
A10/16, 6R/120 for Relative Scuffing Load-Carrying Capacity of High EP Oils.
2014;60:2–17. https://doi.org/10.1016/j.ijfatigue.2012.11.016.
Geneva: International Organization for Standardization; 2004.
[30] Murakami Y, Yokoyama NN, Nagata J. Mechanism of fatigue failure in ultralong life
[7] ISO. ISO 14635-3:2005 - Gears – FZG Test Procedures – Part 3: FZG Test Method A/
regime. Fatigue Fract Eng Mater Struct 2002;25:735–46. https://doi.org/10.1046/j.
2, 8/50 for Relative Scuffing Load-Carrying Capacity and Wear Characteristics of
1460-2695.2002.00576.x.
Semifluid Gear Greases. Geneva: International Organization for Standardization;
[31] Shiozawa K, Lu L. Very high-cycle fatigue behaviour of shot-peened high-carbon-
2005.
chromium bearing steel. Fatigue Fract Eng Mater Struct 2002;25:813–22. https://
[8] Dowling WE, Donlon WT, Copple WB, Chernenkoff RA, Darragh CV. Bending fa-
doi.org/10.1046/j.1460-2695.2002.00567.x.
tigue behavior of carburized gear steels: four-point bend test development and
[32] Winkler KJ, Schurer S, Tobie T, Stahl K. Investigation on the tooth root bending
evaluation. SAE Pap 960977; 1996. doi:10.4271/960977.
strength and the fatigue fracture characteristics of case carburized and shot peened
[9] Phelan PE, Sell DJ, Dowling WE. Bending fatigue behavior of carburized gear steels:
gears of different sizes. Int. Gear Conf., Lyon. 2018. p. 633–44.
planetary gear test development and evaluation. SAE Pap 960978; 1996.
[33] Nelson DV, Long Z. Bending fatigue of carburized steel at very long lives. J Mater
[10] Sieber R. Bending fatigue performance of carburized gear steels. SAE Pap 920533;
Eng Perform 2015;25:220–6. https://doi.org/10.1007/s11665-015-1811-8.
1992. doi:10.4271/920533.
[34] Nakajima M, Tokaji K, Itoga H, Ko HN. Morphology of step-wise S-N curves de-
[11] Benedetti M, Fontanari V, Höhn BR, Oster P, Tobie T. Influence of shot peening on
pending on work-hardened layer and humidity in a high-strength steel. Fatigue
bending tooth fatigue limit of case hardened gears. Int J Fatigue 2002;24:1127–36.
Fract Eng Mater Struct 2003;26:1113–8. https://doi.org/10.1046/j.1460-2695.
https://doi.org/10.1016/S0142-1123(02)00034-8.
2003.00694.x.
[12] Gasparini G, Mariani U, Gorla C, Filippini M, Rosa F. Bending fatigue tests of

11

You might also like