Professional Documents
Culture Documents
Journal of Tribology Copyright © 2010 by ASME OCTOBER 2010, Vol. 132 / 041606-1
冕 冉 冊冉 冊
quency nature of the signal changes with time or is of random +⬁
ⴱ
nature. PWx共t, 兲 = h共兲x t + x t − e−j2d 共4兲
Among different classes of time-frequency distributions −⬁
2 2
共TFDs兲, the energy class of time-frequency transformations is of
practical interest. Usually in energy time-frequency signal distri- where h共兲 is a regular window.
bution, the main target is the total energy of the signal at any 2.3 Smoothed Pseudo-Wigner-Ville Distribution. The
specified time and frequency. This class of TFDs was introduced Wigner–Ville distribution is quadratic in x, which means if x is a
by Cohen in 1966 who utilized bilinear transformations. A draw- sum a + b, the Wigner–Ville distribution of x includes a cross term
back of bilinear TFDs is that they intrinsically possess spurious of 2ab in addition to the quantity 共a2 + b2兲. This expression can be
terms. These cross terms appear when the signal under investiga-
formulated as follows 关39兴:
tion is composed of multiple components. Detailed investigation
of different classes of TFDs and characteristics of each is beyond Wx+y共t, 兲 = Wx共t, 兲 + Wy共t, 兲 + 2R兵Wx,y共t, 兲其 共5兲
the scope of this paper; they are discussed in appropriate detail in
where Wx,y共t , 兲 = 兰−⬁
+⬁
x共t + 共 / 2兲兲y ⴱ共t − 共 / 2兲兲e−j2d
and R
different publications 关32,33兴.
stands for the real part.
2.1 Wigner–Ville Distribution. Wigner–Ville distribution In practice, spurious terms can be considerably reduced by
共WVD兲 is one of the TFDs in Cohen’s class, which is widely used smoothing in time and frequency. The result is the smoothed-
in different engineering disciplines such as speech recognition, pseudo-Wigner–Ville distribution 共SPWVD兲, which is defined by
biological signal processing, acoustics, and earth-quake engineer- the following equation 关32,36,41兴:
冕 冕
ing. The Wigner function was first developed by Wigner in 1932 +⬁ +⬁
for the applications in classical statistical mechanics 关34兴. The SPWVx共t, 兲 = h共兲 g共s − t兲x共s + /2兲xⴱ共s
Wigner distribution was redefined by Ville in 1948 for signal pro- −⬁ −⬁
cessing applications and this new transformation is called
Wigner–Ville distribution 关35兴. − /2兲dse−j2d 共6兲
Wigner–Ville distribution provides a three-dimensional repre- ⴱ
where “ ” is defined as a convolution with respect to time t. The
sentation of signals in time-frequency-amplitude space, but usu- function g共t兲 is the smoothing function in time and h共兲 confines
ally the projection of this three-dimensional representation is
the range of the integral in . Restricting the range in is equiva-
shown in two dimensional time-frequency plane.
lent to smoothing in frequency. The SPWV distribution reduces to
The general formulation for the time-frequency representation
the conventional Wigner–Ville distribution when h共兲 = 1 and
of a signal is given by Cohen 关32兴 and is defined in Eq. 共1兲.
g共t兲 = ␦共t兲. To use the SPWV for practical experiments that use
冕冕冕 冉 冊冉 冊
sampled signals, a discrete form of the SPWV is required as fol-
1 ⴱ lows 关39,40兴:
w共t, 兲 = e−j−j−ju⌽共, 兲s u + s u − dudd
2 2 2 Q/2−1 M/2
兺 兺
2
共1兲 SPWVDs共n,m兲 = 兩h共k兲兩2s共n + k兲sⴱ共n − k兲exp共
⌬t p=−Q/2+1 k=−M/2+1
where s共u兲 is the time representation of the signal and its complex
conjugate is sⴱ, and ⌽共 , 兲 is the kernel 关36兴 based on definition − j2mk/M兲g共n − p兲 共7兲
of which different time-frequency distributions can be obtained. where ⌬t is the length of the data in seconds, M is the size of the
By choosing ⌽共 , 兲 = 1, the WVD is produced, which is defined frequency smoothing window represented by h共k兲, and Q is the
as follows 关37,38兴: size of the time smoothing window which is represented by g共p兲.
2.4 Energy of the Signal in Time-Frequency Domain. By was used for measurement of friction and wear. Figure 3 shows
integration the joint distribution over time or frequency, a one- the schematic of this tribometer. This tribometer can perform re-
dimensional function of frequency or time is obtained. Equations petitive oscillatory tests with a swing angle as low as 30 deg and
共8兲 and 共9兲 show the integration over time and frequency 关36,41兴. frequencies as low as 0.1 Hz. The relative displacement of upper
冕
+⬁ and lower spindles is measured using a levered linear variable
x共t, 兲dt = 兩X共兲兩2 共8兲 differential transformer 共LVDT兲 attached to the lower spindle of
−⬁
which sensitivity is 1.614 V/mm and its accuracy is 2.5
⫻ 10−3 mm.
冕冕
+⬁ +⬁
ings, and Rmean represents the mean radius of the specimen. The
data acquisition system I shown in Fig. 3 can measure coefficient
E= x共t, 兲dtd 共10兲 of friction at maximum rate of 0.125 samples/s. Data acquisitions
−⬁ −⬁
setup consists of a two accelerometers and a four-channel signal
In this paper, we utilize the energy content of the SPWVD as a conditioner. The test rig was fitted with high-resolution piezoelec-
potential feature for fault diagnosis purposes of a solid lubricated tric accelerometers 共model PCB-352B, ceramic shear ICP type
ball bearing. This feature is obtained by two-dimensional integra- with sensitivity of 1000⫾ 10 mV/ g兲 to measure the bearing vi-
tion of matrix SPWV according to the time and frequency. As the bration signals. One of the accelerometers is mounted on the
severity of the faults expands, the energy content of the vibration lower specimen holder and the other is mounted on the upper
signal increases significantly, providing useful information on the spindle of machine. The data acquisition board used for the vibra-
energy content in time-frequency plane. The time-frequency tool-
box for MATLAB 关42兴 is used for obtaining the energy feature from
SPWV.
tion measuring is a DT300 family DAQ board with 16 bit incorporate both of these elements is needed. In this research,
resolution. MoS2-coated thrust ball bearings are tested under varying loads
Five series of experiments are conducted using the tribometer and types of motion.
to assess the behavior of the coating over extended period of time. In what follows, the results of the calculation of the energy of
The operational parameters include types of motion 共unidirec- the smoothed-pseudo-Wigner–Ville time-frequency representation
tional or oscillatory兲, frequency of oscillation, load, and coating are used for different types of experiments. The bearing is ex-
thickness, as summarized in Table 2. The tangential sliding speed posed to a normal load of 177 N and the frequency of oscillation
共speed of the center of the balls兲 for cases 1–3 is 6.02 is 0.5 Hz. The MoS2 coating of thrust bearings used in this ex-
⫻ 10−3 m / s, and for cases 4 and 5 it is 12.04⫻ 10−3 m / s and periment is 9650 Å thick. Figure 4 shows the coating surface at
18.06⫻ 10−3 m / s, respectively. Assuming that each ball experi- different stages of its life: Fig. 4共a兲 is the fresh coating surface and
ences one-twelfth of the total load and using the Hertzian contact Fig. 4共b兲 is the coating after 360 min of operation while the bear-
theory, the nominal pressures between steel balls on a plane are ing was healthy. Figure 4共c兲 shows the lower raceway when the
23.6 MPa for 355 N load and 18.8 MPa for 177 N load.
coating is depleted.
3.2 Time-Frequency Analysis for Coating Damage MoS2 wear occurs due to different mechanisms such as fatigue,
Detection. Statistical time domain vibration analysis alone is not adhesive, oxidative, and plowing As a result of wear, MoS2 debris
sufficient to detect ball bearings failure accurately and reliably. is produced. A portion of debris creates a transfer film by adhering
Since coating wear simultaneously affects both the time and the to the balls, which provide some useful lubrication. The rest re-
frequency representation of the vibration signal, a method that can mains on the raceways for a period of time but is eventually
Fig. 4 Three stages of coating life. „a… Fresh MoS2 coated surface. „b… Coated surface after 360 min
while bearing is operating in healthy regime. „c… Failure of the coating. The arrows show the debris
ejected from the raceway and accumulated on the sides of the raceway.
pushed out of the contact area. The right-hand-side arrows on Fig. period 共phase II兲, and the failure phases 共phase III兲. The duration
4共c兲 show the accumulated debris on the sides of the raceway. As of each phase is depended on the operating parameters as well as
the experiment continues, the debris is ejected out of the raceway the experiment conditions such as load, speed, type of motion
and, upon depletion of coating, eventually the balls come into 共oscillatory or unidirectional兲, coating thickness, and humidity.
contact with the race metal substrate, triggering the initiation of Phase I provides satisfactory operation for sensitive mechanisms
the failure phase. that demand accurate control of friction, while in phase II the
Continuous recording of data during the experiments 共duration integrity of the solid lubricant is compromised. Phase III repre-
time of 1500–6000 min兲 generates huge data files that are not sents the failure of the coating and the performance of the device
conducive to simulation of a time-frequency distribution on a per- is no longer determined by the lubrication characteristics of MoS2.
sonal computer. The PC on which the program is run has an Intel Figure 5共b兲 shows the coefficient of friction as a function of
Pentium III Xeon processor 2.49 GHz and a 3 Gbytes of RAM. time. The test results reveal that under the conditions tested, the
During each experiment, the vibration data are recorded in 10 min friction remains fairly constant for about 800 min of operation
intervals and saved as a MATLAB file data format. Then, the first after which it begins to experience a rapid rise. The tendency of
2048 points of a 10 min signal is selected to calculate the SPWV the coefficient of friction to rise is indicative of a change in the
distribution. The result is a 2048⫻ 2048 matrix of the SPWV dis- surface characteristic and MoS2 debris generation. As the test con-
tribution. This is the maximum number of points to calculate
tinues, this MoS2 debris generation accelerates and leads to a sig-
SPWV-TFD that the PC could handle because of memory issues.
nificant increase in coefficient of friction. After around 980 min,
However, the results with this amount of data for each calculation
the friction tends to decrease. The reason for this phenomenon can
are satisfying. For instance, the 6000 min experiment yields 600
be explained as a result of ejection of MoS2 debris from the
vibration data files, each with the duration of 10 min. The 2048
raceways.
first points of each file is chosen for calculating the SPWV. The
next step is to determine the energy in time-frequency plane for As a result of wear and MoS2 debris generation, higher values
the achieved TFD. This procedure is repeated for all of the 600 for coefficients of friction are registered as a natural reaction to
vibration files. Finally, 600 points are generated for the 6000 min surfaces of the raceways becoming rougher. Values of the energy
experiment that can be plotted to examine the variation of the feature in TF plane are demonstrated in the Fig. 5共a兲. The first 800
energy feature as the test continues. Comparison of this feature min of this test is considered as the healthy part of the coating life
with the variations of coefficients of friction provides information 共phase I兲 after which the coating starts to degrade. Compared to
on the degradation of the MoS2 coating. the healthy period, the coefficient of friction rises from a value of
around 0.05 to almost 0.2. This period, which is called developing
4 Establishing the Feature Potency With Experimen- damage stage 共phase II兲, lasts for approximately 180 min. At the
tal Case Studies end of this period, the top and bottom raceways are nearly de-
pleted of solid lubricant. This period of time is considered to be
4.1 Case 1: Unidirectional Rolling Motion. The aim of this the failure phase of the MoS2 coating 共phase III兲. Figure 6 shows
test is to investigate the behavior of the MoS2 coating under the the depleted raceway in failure phase where the worn area can be
unidirectional load. This test is performed for 3000 min under a easily seen. The white arrows show debris accumulation on the
load of 355 N. The thickness of the MoS2 coating in this case is sides of the raceway.
9650 Å. The vibration signal is sampled by 2.5 kHz of frequency. Turning our attention to the Fig. 5共a兲, we note that during the
Three broad phases of the operation as identified in Figs. 5共a兲 first 830 min of operation the energy feature in TF plane is nearly
and 5共b兲 are the healthy period 共phase I兲, the developing damage constant. It shows a very distinct increase in its value compared to
value of the energy in SPWV time-frequency plane changes from sifies around 700 min. The arrows in Fig. 9 highlight distinct
roughly 0.42 关V2兴 in 200 min to a value of 640.4 关V2兴 after 220 intervals in which coefficient of friction show rising local slopes.
min, thus clearly signifying a step change in the performance of A rise in coefficient of friction represents an increased resistance
MoS2. Subsequently, the bearing enters its failure phase and its against rolling motion, as well as the consequent debris
behavior is no longer governed by MoS2. In contrast, the time generation/ejection. These alterations in frictional behavior of the
domain vibration signal in Fig. 10 shows sharp pulses in around coated system are due to the occurrence of wear on the layers of
200 min; this behavior continues and even to some extent inten- the MoS2 remaining on the raceways. The trend is similar to what
happens while wear first appears in the developing damage phase. also increase these two factors. The above-mentioned conse-
In each of these intervals, the alterations in surface characteristics quences are observable in case 3 共Figs. 11共a兲 and 11共b兲兲 after
leaves a distinct vibration signature with respect to the healthy entering the failure stage. The coefficients of friction plot in case
period and consequently the energy feature increases in a reaction 3 shows multiples of consecutive rises and falls after its first in-
to such changes. crease in around 1840 min. As mentioned earlier, the rises in the
In cases 3–5, the oscillation frequency is varied, and the capa- coefficient of friction values indicates the alteration in lubrication
bility of detection of the damage by the energy feature is assessed. regime, which is due to wear, MoS2 debris generation, and effec-
The results are presented in Sec. 4.3. tively rougher surfaces. The load on the coated thrust ball bearing
4.3 Cases 3–5: Oscillatory Motion With Different Loads in case 3 is one-half of the load applied in previous cases. This
and Frequencies. The following results pertain to three other os- results in a multistage wear occurrence, which means that the
cillatory cases under different loads and frequency tests. In these solid lubricant undergoes multiple stages of damage compared to
cases, similar to those identified previously, the damage evolution the one-stage wear, which is observed in case 2 共Figs. 9共a兲 and
is recognized by a rise in coefficient of friction as well as energy 9共b兲兲.
feature obtained from time-frequency representation of the signal Table 3 tabulates the phases in case 3 where the damage occurs
and variations in health changes is apparent. Figures 11, 12, and and gives the corresponding coefficient of friction and energy fea-
14 show the results under different frequencies of oscillation and ture. Arrows in Fig. 11共b兲 show the different stages for damage
load. After an approximately 1840 min 共Figs. 11共a兲 and 11共b兲兲, occurrence and rising slope in the coefficient of friction. Five
simultaneously with the rising in coefficient of friction, the energy phases are identified after a healthy period, which are shown in
feature starts rising and its value reaches 1.63 关V2兴 after 1920 min. Fig. 11共a兲. Phase II is the first time wear appears on MoS2 surface,
Such an increase in both plots represents the change in health which is called damage developing phase. Phases III, IV, V, and
phase of the coating and transition from healthy period 共phase I兲 VI indicate alteration in surface characteristics inside the failure
to the developing damage phase 共phase II兲. The process continues phase of the MoS2. After around 4000 min, the solid lubricant is
until in 2060 min of operation the energy feature reaches 18.71 nearly worn and incapable of providing lubrication. Metal to
关V2兴; this is also a representative of the developed damage and metal contact and abrasive metal debris exist in this phase of the
identified as phase II. At this stage performance of MoS2 deviates operation.
from its normal working condition, with quite distinguishable In case 4 共Figs. 12共a兲 and 12共b兲兲, a healthy period of 200 min is
changes that can be seen in coefficient of friction. During a de- recognizable after which the developing damage phase forms. A
veloping damage phase, layers of MoS2 are detached and the rise in coefficient of friction at this time is indicative of the ap-
thickness of the coating is reduced in some areas of the top and pearance of damage. At the 290 min of the operation while ap-
bottom raceways. Such regions are prone to more wear and dam- proximately the coefficient of friction has reached its maximum
age. Figure 13 shows the worn surface of MoS2 coating in which value, the energy feature reaches a value of 11.3 关V2兴. By com-
traces of MoS2 debris and wear can be observed. “Scratchy sur- paring cases 3 and 4, it is noted that increasing the load and
face” is indicative of the reduced thickness of the MoS2. frequency of oscillation decreases the life of the coating signifi-
The level of the energy in time-frequency plane, as well as the cantly. However, in both cases multiple failure occurrences are
coefficient of friction, increases when surfaces become rougher as detected. Arrows in Fig. 12共b兲 show an increasing trend of the
a consequence of wear or MoS2 debris generation, which tend to coefficient of friction, indicative of the occurrence of local wear
impede the smooth rolling of the balls. Metal to metal contact can and MoS2 debris generation. For each of these intervals within
which a failure occurs, a significant increase in energy of the cient of friction. As it is depicted in Fig. 14共b兲, the rise in coeffi-
SPWV is distinguishable in Fig. 12共a兲. Phase I is identified as the cient of friction has a lower slope and the damage evolution has a
healthy period of the solid lubricant in Figs. 12共a兲 and 12共b兲 and slower developing rate. This trend of damage evolution might be
phase II is developing damage phase after which MoS2 enters the due to the increase in frequency of oscillatory motion. It is ob-
failure stage of its life. Intervals III–VII are distinguished damage served in Fig. 14共b兲 that the coefficient of friction has multiple
forming stages, which are located in failure phase. Case 5 is per- stages of slow peaks and valleys, taking place approximately
formed with an increased frequency of oscillation equals 1.5 Hz. around 230 min, 470 min, 810 min, and 1090 min. It is believed
In case 5 共Fig. 14兲, it is shown that applying higher frequency that in these four stages, a considerable amount of the coating
of oscillation, which is 1.5 Hz, causes a different trend in coeffi- leaves the raceways. Monitoring the energy feature in these time
intervals show distinct values compared to that of other time pe-
riods. However, approximately 130 min of healthy period is quite
distinguishable and is labeled as phase I in Figs. 14共a兲 and 14共b兲.
What is promising with this test is that although the MoS2 shows
a different behavior as it is undergoing wear and damage, the
energy feature can represent the damage evolution coherently. Af-
ter passing the healthy period and entering the failure phase, in
different time intervals, debris generation continues and MoS2
continues to leave the raceways, as it is illustrated in Figs. 14共a兲
and 14共b兲. Compared to the previous case, healthy period duration
decreases because of applying the higher frequency of oscillation.
In addition to the investigations of the potency of the energy
feature calculated from SPWV, the effect of the operational pa-
rameters on the coefficient of friction 共COF兲 during the healthy
period is studied. Table 4 shows the range of the coefficients of
friction during healthy period of the five cases studied.
Fig. 13 Reduced thickness of MoS2 coating
The results of cases 2, 4, and 5 can be used to analyze the effect
Table 3 Phases in which damage occurs and corresponding of the sliding speed on coefficient of friction since they are per-
coefficient of friction and energy in TF plane „Case 3… formed under the same load and type of motion. As it can be
inferred from the Table 4, increasing in the frequency of oscilla-
Coefficient of friction Energy of tion, which is equivalent to increasing in the sliding speed, causes
Phases in which in each interval SPWV value a reduction in the level of the registered coefficients of friction.
damage occurs 关共min兲–共max兲兴 共V2兲 The minimum, maximum, an approximation of the mostly re-
corded coefficient of friction, as well as the value of the coeffi-
Phase II 关0.023–0.11兴 18.71
cient of friction at the end of the healthy period, decrease as the
Phase III 关0.023–0.11兴 29.26
Phase IV 关0.03–0.16兴 23.29 sliding speed boosts in cases 2, 4, and 5. Case 3, which has the
Phase V 关0.024–0.123兴 8.167 lowest applied load, is of the longest healthy period. This repre-
Phase VI 关0.027–0.074兴 53.2 sents an adverse effect of the higher load in healthy life of the
coating. Additionally, the length of the healthy period in case 1,
which is performed in a unidirectional type of motion, is higher coefficients of friction of the MoS2 also are measured that show
than those of cases 2, 4, and 5 that correspond to an oscillatory the appearance of the faults on the coating. Both of methods show
type motion but with an equivalent load. This reveals that in an similar result in detecting the wear occurred on the raceways. In
equal loading condition, a unidirectional type of test results in a cases 1 and 2, after the healthy period, solid lubricant enters the
longer healthy period in comparison to that of an oscillatory case. developing damage and failure appears in one step. This can be
recognized by dominant rises in coefficient of friction, as well as
5 Conclusions energy feature. In cases 3–5 the solid lubricant after passing the
healthy period undergo several developing damage and failure
In this research, the feasibility of extracting a feature from occurrences. The multiple failure occurrences are detected by sev-
time-frequency analysis to detect the wear of the MoS2 coating on eral consecutive rises in coefficient of friction, as well as in en-
thrust ball bearings is investigated. SPWV representation is se- ergy feature. These results support the potency of this feature in
lected for calculating the TF distribution. The energy of the signal reliable diagnosis of the MoS2 operating phases and in recogniz-
in TF plane is employed successfully as a monitoring feature. The ing stages in which the lubrication regime in solid lubricant de-
results of the five sets of experiments establish the potential of the grade and undergo undesirable variations.
proposed method for detection of the wear of the coating, as well The energy feature demonstrated independency from the oper-
as the changes in lubrication regime of the solid lubricant. Three ating condition, such as type of motion, frequency of oscillation,
stages of the coating life are recognized during the experiment and loads, and detects the stage at which the coating starts leaving
from start to failure. These stages are called the healthy period, the raceways. This behavior makes the feature useful for pattern
developing damage, and failure. The duration of the each depends recognition and automatic fault diagnosis software. While in many
highly on the load, type of motion, and the coating thickness. In tribology applications real-time monitoring the coefficient of fric-
order to show the potency of the feature in damage detection, tion of the coated ball bearings may not be possible, monitoring
such feature extracted from vibration signal of the coated ball
bearing provides an attractive alternative for coating fault diagno-
Table 4 Summary of the measured coefficient of friction dur- sis purposes.
ing the healthy period. The second column expresses the min
and max of the coefficient of friction in healthy period. The
third column provides an approximation of coefficient of fric- References
tion during the healthy period and the last column gives the 关1兴 Pritchard, C., and Midgley, J. W., 1969, “Effect of Humidity on Friction and
values of the friction at the transition point. Life of Unbonded Molybdenum Disulphide Films,” Wear, 13共1兲, pp. 39–50.
关2兴 Black, A. L., Dunster, R. W., and Sanders, J. V., 1966, “Molybdenum Disul-
Approximation phide Deposits—Their Formation and Characteristics on Steel Balls,” Wear,
COF in healthy of the mean COF at 9共6兲, pp. 462–476.
period COF in transition to the 关3兴 Palmer, E. B., 1960, “Jet and Space Age Lubrication,” Ind. Lubr. Tribol.,
Case 关共min兲⫺共max兲兴 healthy period second phase 12共2兲, pp. 12–17.
关4兴 Campbell, M. E., Loser, J. B., and Sneegas, E. S., 1966, “Solid Lubricants,”
NASA Report No. SP-5059.
Case 1 关0.010–0.024兴 0.017 0.036
关5兴 Kim, K., and Korsunsky, A. M., 2008, “Exponential Evolution Law of Fretting
Case 2 关0.012–0.040兴 0.028 0.040 Wear Damage in Low-Friction Coatings for Aerospace Components,” Surf.
Case 3 关0.020–0.054兴 0.022 0.064 Coat. Technol., 202共24兲, pp. 5838–5846.
Case 4 关0.010–0.026兴 0.012 0.020 关6兴 Zhu, M. H., and Zhou, Z. R., 2001, “An Investigation of Molybdenum Disul-
Case 5 关0.007–0.017兴 0.008 0.013 fide Bonded Solid Lubricant Coatings in Fretting Conditions,” Surf. Coat.
Technol., 141共2–3兲, pp. 240–245.