Академический Документы
Профессиональный Документы
Культура Документы
0890-6955(95)00058-5
Abstract--This paper presents a study on monitoring tool wear and failure in drilling using vibration signature
analysis techniques. Discriminant features, which are sensitive to drill wear and breakage, were developed
in both time and frequency domains. These features were found to be relatively insensitive to cutting
conditions, and sensor location. In the time domain, a monitoring feature based on calculating the kurtosis
value of both tlae transverse and thrust vibrations, was found to be rather effective for on-line detection of
drill breakage. On the other hand, in the frequency domain, a cepstrum ratio, derived from the spectra of
the vibrations monitored in both directions, was also found effective in detecting breakage events. The effect
of different types of wear on the vibration power spectra, in both the transverse and the thrust directions,
was also investigated. A signature feature, namely the instantaneous ratio of the absolute mean value
(RAMVI), was developed in this study and used as a threshold for controlled capture of the vibration signal.
The ability of the monitoring features to detect drill wear and breakage was verified experimentally. The
drilling tests were performed using 3 and 6 mm diameter high speed steel twist drills, and cast iron workpieces.
The results confirmed the effectiveness and robustness of the proposed monitoring features. Copyright ©
1996 Elsevier Science Ltd
1. INTRODUCTION
tlntelligent Machines and Manufacture Research Center, Mechanical Engineering Department, McMaster
University, 1280 Main street West, Hamilton, Ontario, Canada LSS 4L7.
687
688 T . I . El-Wardany et al.
15
14
13
12
o 11
Fap
"" 9
~ s
~ 7
,~ 6
= 5
Z 4
0
Load Image Proximity Work Touch Torque
amperage Processing Switch Sensor Sensing
Vibration Ring Tool Touch Acoustic Size
Sensor Sensor Sensor Emission Measuring
Fig. 1. Different monitoring methods for detecting edge chipping, tool fracture, tool wear and poor hole
quality [4].
system establishes a "tare" torque from the no load condition, and subtracts it from
the total torque calculated. The disadvantage of this monitoring technique is the
measurement bandwidth which is limited by the motor inertia. Also research performed
by Matsushima et al. [6] to detect tool failure in turning by measuring the power
reported a delay of 100-300 msec. An eddy current sensor installed on the drill shank
was used by Birnksmire [7] to measure the average torque, and to detect tool fracture
based on analyzing the dynamic component of the signal. This technique was applied
successfully on drills of 10 mm diameter.
Acoustic emission (AE) was reported to be approximately 33% successful when
used to monitor drill failure [4]. Saijo and Noh [8] monitored drill fracture using AE.
In their work, a large amplitude (spike) in the AE signal was found to be indicative
of drill fracture. The amplitude of the AE spike was strongly dependent on the fractured
area of the drill and its position [9]. The major difficulty in AE monitoring is the
development of appropriate filtering techniques and algorithms to separate useful signals
from background noise generated during the drilling process. In addition, problems
associated with transducer mounting and path dependency of the signal have also been
reported [4].
Monitoring thrust force-torque to predict drill failure has been widely used and
reported by several investigators [10-12]. The advantage of this technique is the
sensitivity of the thrust force-torque signals to changes in the drill conditions. Thangaraj
and Wright [12] used the rate of change of thrust force in predicting drill failure. This
rate of change was reported in their study to be independent of the cutting conditions
and the drill, geometry, which facilitate the use of a prespecified threshold to detect
drill failure.
Sensor fusion methods were used by Liu and Wu [13] for monitoring drill wear. In
this investigation both the thrust force and axial acceleration signals were used. A
fuzzy pattern recognition technique based on measuring the thrust force and torque
was used to monitor drill wear [14]. Subramanian and Cook [11] emphasized that drill
condition monitoring using the thrust force and/or torque was effective only when a
close tolerance on the workpiece hardness was maintained (about 5%). Neural networks
were proposed for the detection of tool failure in micro-drilling operations [15]. This
was done by encoding thrust force signals with wavelet transformations and classifying
Vibration Signature Analysis 689
the estimated coefficients using adaptive resonance theory. This technique was rather
difficult to apply in situations involving small signal-noise ratios.
Vibration monitoring techniques applied to the detection of drill failure have been
reported by several investigators [2, 3, 16-18]. The advantages of these techniques
include ease of implementation and the fact that no modifications to the machine tool
or the workpiece fixture are required. However, the disadvantages reported in the
literature include dependency of the vibration signals on workpiece material, cutting
conditions, and machine structure. Clearly, additional research is required to develop
practical vibration monitoring techniques which are sensitive to tool conditions but
relatively insensitive to cutting conditions, sensor location, etc.
The present work deals with the development of vibration-based monitoring methods
for detecting breakage of small size drills (3 mm diameter) and wear of larger size
drills (6 mm diameter). The characteristics of the vibration signatures were examined
in both time and frequency domains. Signature features were developed and were
shown to be effective in detecting drill wear and breakage, while insensitive to variations
in cutting conditions. The effectiveness of the proposed monitoring features was verified
from drilling experiments using high speed steel twist drills and cast iron workpieces.
2. CHARACTERISTICS OF VIBRATION SIGNATURES IN DRILLING
Corner Wear
"0
= 0,5
o. 0
E
C
-0,5
O
~ -1.5
-2 i. i i
0 0 .9 ' 0:4 0:6 '
0.8
Drilling Time (sec)
a) Sharp Drill
~'1.5 Hole # 5
O
--I
0.5
~. ......... , , L _ tilL,.,
~. 0
oq.d -1
¢=
.~ ..1.5
"2' J i I i
o o', 0'.4 o.0 o.0
Drilling Time (seo)
2
Hole # 6 Breakage
~ 1.5
Event
1
0.5
}o
-11.5
-1.5
-2
0 o'.2 ' 0'.4 ' o'.6 ' o'.e
Drilling Time (sec)
c)Drill Breakage
Fig, 3. Vibration measured in Z direction.
Hole#1
0.5
0
~C , 5
0
~ -1.5
-2 o ' o~ ' 0'.4 ' o'.e ' o'.8
Ddlllng Tlme (sec)
a) Sharp Drill
2 I
-~1.5 ¸. I
~ o.s!-
{ol I
~.o.5 ~- !
C :
~ "1i" !
~-1.5 -
"2"
0" 0.2 0.4 0.6 0.8
Orllllng Time (sec)
1.5-
.a,
~ 0,5
b-i
~1"5'
N" ~2
0 0.2 0.4 0.6 0.8
Dr.llng Time (sec)
c)Drill Breakage
Fig. 4. Vibration measured in Y direction.
~ 0.15-
0>
6
5000
a) Z Direction
........-''"~"...
'...,. . . .
.....'""
i : ''""
4.
D
.......... i ! .......
1-
0:
6 "
5 .° "",,., . " "'.o .. """ ... "~'~ i
5000
"°~'0,,~_ 3 ~ ~....~"~.. ~ 3000
U~OO~~ 2 " l ' 1000 2000 "G~ t~z~'
e,S' 1 0 t:B¢---O~3Et~
b) Y Direction
crater wear, and chisel edge wear. These are illustrated in Fig. 6 together with chipping
of the lip. The dominant types of wear which result in drill failure and breakage are:
chisel wear, outer corner wear, flank wear and margin wear. The effect of each of
these four types of wear on the power spectra of the vibration signals will be described
and discussed below.
2.2.1. Vibration spectra in the transverse direction. Figures 7(a)-(h) show examples
of the power spectra of the vibration signals generated in the Y direction during drilling
with various levels of different types of induced wear. The tests were performed at a
cutting speed (V) = 12 m/min, f = 100 mm/min, D = 6 mm and L = 30 mm. Figures
7(a) and (b) show the effect of chisel wear on the vibration power spectra. It can be
seen that for a chisel wear of 0.225 mm [Fig. 7(a)], the power spectrum has two peaks
at frequencies of approximately 3.8 and 4.8 kHz, neglecting the peak existing at a
frequency of 720 Hz which was related to noise from the machine motor. The first
peak was related to one of the spindle modes in the Y direction, while the second one
was related to a workpiece mode in the Z direction. The first peak was damped out
with the increase in chisel wear up to 0.3 mm. However, increasing the amount of
chisel wear to 0.608 mm [Fig. 7(b)] shows an increase in the magnitude of the spectrum
in the frequency range of 4.5-5.8 kHz and a peak at approximately 5.3 kHz. The
'i
amplitude of this peak increased gradually with the progress of wear. As shown in
Lip
Outer corner
c) Chisel Edge Wear f) Chipping at Lip
0
i = ! ~ ! 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6 7
Frequency (kHz) Frequency (kHz)
c) Corner Wear = 0.49 mm d) Corner Wear = 0.98 m m
~'~ 0.0015
0.002 ]- i
i i 1
i
[
]
~ " 0.0015
;>
°°L !1 1
0.001 t ............! ........... ~ .............r . . . . . . . ~-~---t----I "~" 0.001
0 0
2.! ~_
J I
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Frequency(kHz) Freauency (kHz)
e) Margin Wear = 0.28 mm f) Margin Wear = 0.824 m m
0.005 0.005 ! ~ ~ i t
~, 0.004 ...........................-..............~............~.............~........... 4.......... ~ o.oo4 ..............~...............,...............t ............~........-r..... -~........
--o.oo~
.......... - ............ !.............. !~............ ! .............. -~ . . . . . . ! ...........
i
~ o 0 . 002 ............. : ............ ~ ............. '- ............ ~ ......... i i
generated during drilling with different values of artificial wear induced on the flank
of the tool. As can be seen in these figures, flank wear excites one mode only with a
peak at approximately 3.9 kHz, which corresponds to one of the natural frequencies
of the tool holder in the Y direction.
The above results show clearly that the vibration signature is rather sensitive to
flank, corner, and margin wear.
2.2.2. Vibration spectra in the thrust direction. Figures 8(a) and (b) represent the
power spectra in the Z direction obtained at different levels of chisel wear induced on
the drill. The cutting conditions were the same as those used for the tests in the Y
direction. The trends displayed are clearly similar to those found in the Y direction.
However, the vibration spectra resulting from corner wear in the Z direction are
different from those obtained in the Y direction. As shown in Fig. 8(c), two regions
with frequency ranges of 3-3.5 and 3.6-3.9 kHz were excited during drilling with a
......
,'~.~
I
.... ~_....~ ...... ,
I
'LI
I I
o.~
~ .0.002
0.11111 ='o.ool _~1 t ~k -~ I
0
1 2 3 4 5 6 0 1 2 3 4 5 6 7
0
Frequency (kHz) Frequency (ld-lz)
_.j~. . . . .
i
i
....... " - i - - - I
I
~
t ....
~
!__
- = -V--1
Jf - - - ~ - - - I I
! i
I "-" 0.003 J
!=
t-- t ........... 1. . . . . "1 .....
t. " ' ~ ~ | t J =
o 0 002 / I 1 I , ~ ~-----I
o ,l .... i . . . . L I i - - t .... r-i 0.001 - - 4 -
0 L~..-! _^
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6
Frequency (kHz) Frequency (Id-lz)
e) M a r g i n W e a r = 0.28 m m f) M a r g i n W e a r = 0.824 m m
0.005 |' . ; i ~
,-., 0.005
"-" o.~, ..........~i....... ~~........ ~..... t-----~= ...... l I, ~, 0.004 1 1 1 i /
----~--- ,-----~.----I,-----1~
3"- o0o~t ........ ~............i ......... ! ...... i ....... ~ ...... "--.1 ' ~ 0.003
, ........i I i! 1 ~ o.~ | 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0
~ 0.~1-
^t,,, , ~ I .ui i i
u - : . : -: ~ . i !
0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6
Frequency (kHz) Frequency (kHz)
g) A v e r a g e F l a n k W e a r = 0.1 m m h) A v e r a g e F l a n k w e a r = 0.62 m m
Fig. 8. Power spectra calculated in the Z direction for artificial wear tests.
Vibration Signature Analysis 697
tool that had a comer wear of approximately 0.45 mm. Figure 8(d) indicates that the
power amplitude in these two regions increased with the progress of wear. The increase
in corner wear causes the amplitude of vibrations in the frequency range of 3.0-3.5 kHz
to increase gradually. An almost identical trend was seen in the signal produced in the
Z direction by the various values of margin wear [shown in Figs 8(e) and (f)].
The data collected to study the effect of the flank wear are shown in Figs 8(g) and
(h). As in the Y direction, the vibration amplitude increased for higher values of flank
wear. Howew:r several modes were excited in the Z direction. These modes occur in
frequency ranges between 3.5 and 4.4 kHz, 4.5 and 5.1 kHz, and 5.3 and 5.7 kHz.
In general the vibration signals measured in the Z direction were predominantly
affected by corner wear and margin wear. Flank wear and chisel wear have similar,
but lesser effects on the vibration spectra. The results show clearly that the vibration
spectra can be used to identify worn drills. Various types of information may be
extracted from the analysis of the vibration signals in both the time and the frequency
domain.
This section proposes monitoring features in both the time and frequency domains
which are based on the vibration signature characteristics as revealed above.
1 '=." (x,
K = ~/xz~, (1)
i=1
where xi is the instantaneous amplitude of the vibration signal, :t is the average vibration
amplitude, 0-2 is the variance of the signal and N is the number of samples.
Computer simulation runs were performed to examine the sensitivity of the kurtosis
value to sudden impulses (or spikes). Figure 9 shows that the kurtosis value for a sine
wave is 1.5. "]?his value is constant irrespective of the signal amplitude or frequency.
The kurtosis value varies significantly when a single spike is introduced to the simulated
sine wave as ,~hown in Fig. 10. In this figure, the higher the ratio of A/B, the larger
the kurtosis wtlue. Increasing the number of spikes introduced in the simulation, while
keeping the ratio A/B constant, reduces the kurtosis value as shown in Fig. 11.
The above simulations show clearly that the kurtosis value is rather sensitive to the
occurrence of a single spike or an impulse in the vibration signal. However it is also
sensitive to the generation of a number of these spikes, which tends to reduce its
magnitude as compared to the situation for a single spike.
40,
No. Sl~klm ~ None
- 1.499985
30-
20
~ o
~-10
-20
I I I I i I
0 1 2 3
Tlme (seconds)
4O
3O
2O
~10
0
'13
= 0
n
~-10
-3O
0 1 2 3 4
Time (seconds)
where the power' spectrum of the time series fx(t) is given by: •
N is the value of nAt, where At is the sampling interval, and n varies from 0 to 1023.
The variable f represents the frequency KAf (where Af is the line spacing in the
Vibration Signature Analysis 699
4O
L~ -,-m I
20'
I A
10
O
"O 1
O.
~,-10
-20
-30
-40
3O
i
I
~,-10
, a i ,, & I I I
0 2 3 4
Time (seconds)
frequency spectrum) and in principle K also varies from 0 to 1023. The unit of the
cepstra abscissa (Quefrency) is 1/I-lz or second, and the position of the first peak
corresponds to the reciprocal of the spacing of the side bands or harmonics in the
power spectrum.
Figure 12 shows examples of power spectra and cepstra calculated from vibration
signals measured during a normal drilling process and during drill breakage. The
cutting conditions used in this set of tests were N = 3000 rev/min, f = 750 mm/min,
L = 15 mm and D = 3 mm. Spikes generated after drill breakage, as described earlier,
create a series of side bands in the power spectrum, as shown in Fig. 12(c). The
spacings between these side bands in the power spectrum will be represented by a
700 T. I. EI-Wardany et al.
40
3O
2O
+o
t.1: i" 0
.10
,2O
i i i i i ,
4O
o ; 2 ; 4
Time (.eoonds) Time (second.)
3O
2O
i
'_.j
-20
J i i i L J i i
o ' + ' +, s 4 - 0 1 2 3 4
TIrne (~onds) Time (seconds)
c) Number of spikes = 2, kurtosis = 33.95 d) Number of spikes = 20, kurtosis : 6.94
series of peaks in the corresponding cepstrum, as shown in Fig. 12(d). The cepstra
calculated according to equation (2) vary considerably from one hole to another.
However, the cepstra calculated for the signal measured during drill breakage [see Fig.
12(d)] showed consistently a peak at a Quefrency related to the time of one spindle
revolution. This phenomenon was observed in all drilling tests performed for detecting
breakage in this investigation.
Monitoring the existence of a peak in the cepstra corresponding to the time of one
spindle revolution can be used as an index for detecting drill breakage. However, a
significant improvement to the sensitivity of this technique can be obtained by calculat-
ing the ratio of the cepstra generated in the Y and Z directions. This technique will
be discussed further in section 5.
3.2.2. Monitoring drill wear. The vibration signals measured during actual drilling
were examined and compared to those determined from the controlled wear signals.
The vibration power spectra measured in both the Y and Z directions, at different
states of drill wear, are shown in Figs 13(a) and (b), respectively. Drilling was performed
at cutting speed ( V ) = 12 m/rain, feed rate ( f ) = 100mm/min, drill diameter
(D) = 6 mm and for a length (L) = 30 mm.
When the drill used was sharp, none of the significant vibration modes were excited.
After six holes were drilled, the tool was examined under an optical microscope. Plastic
deformation on the drill flank as well as chisel wear were found to exist. The peaks
in the spectra occurred at 3.9 and 5.3 kHz. Recalling the artificial wear testing, it can
be seen that the peak frequency of 3.9 kHz, in the Y direction, corresponds to the
mode that is excited by flank wear. Similarly, the 5.3 kHz frequency is one of the
modes that is excited by chisel wear. Therefore, it can be concluded that these peaks
were a direct result of the amount of wear on the drill.
These types of wear continued until 21 holes were drilled. At this point, it was
observed that corner wear began to develop on the tool. This was accompanied by a
Vibration SignatUre Analysis 701
o.oo,OO,o o
'
~0
.
!
.
1ooo
.
1~
.
2000
I
.
2so0
.
300o
.
|
.
-_
3500
.i _~1_i
4o00 ~ ~o
Frequency (Hz)
a) Power Spectrum for Vibration Signal Measured During Drilling Hole No. 1
().0~ | i
n,"
~ 0.04
~0.0~
l).0~
11.01
O
5 10 15 20 ?_5
Quefrency (msec)
#.
0 1 1500 2000 2500 300O
Frequency (Hz)
c) Power Spectrum for Vibration Signal Measured During Drilling Hole No. 6
O.05 i i t i i
E"
0 0-04
After breakage
~.~-,0.03
O.02
I.-
03
0.01
O
o; 5 10 15 20 25
Quefrency (msec)
rather small peak, observed in the power spectrum calculated in the thrust direction,
at a frequency of approximately 3.2 kHz. This frequency was approximately the same
frequency excited by the artificial corner wear.
This pattern continued through the drilling process until hole number 32¢ when a
very large peak occurred at a frequency of 4.5 kHz. The tool was removed from the
machine and examined under the optical microscope, It was found that the margin
wear had increased on the tool: The frequency 4.5 kHz was within the region that was
excited with artificial margin wear.
•Starting from hole number 60, the vibration magnitude increased in the frequency
range of 2.4 kHz to 5.8 kHz. The next significant change in the vibration signals
occurred at hole number 128. As can be seen from the power spectrum corresponding
to this hole, several peaks existed in both the Y and Z spectra. These peaks were
702 T . I . El-Wardany et al.
a) Y direction
0.1~
3"
L.
o
o
0.08 -
0.06 -
0.04,
0.02,
211
O:
i!!iiiii!!iii!i!!!iiii!ilii 0 0 f ~ 6 0 0 0
b) Z direction
Fig. 13. Power spectra calculated for vibration signals measured during actual drilling. V = 12 m min -1,
f = 100 mm min -1, D --- 6 mm, and L = 30 mm.
similar to those excited by the different types of artificial wear. Examination of the
tool showed an increase in all the wear types; flank, chisel, comer and margin wear.
The trend found at hole 128 continued, with only the amplitude increasing until hole
number 151. In this case, a peak occurred at a frequency of 4.8 kHz. Examining the
tool revealed a very large built up edge on the chisel edge. No significant changes in
the power spectra were noticed from those corresponding to hole number 128 until
hole number 211, when several peaks of large amplitude were found to exist.
In drilling holes numbered 211-217, it was observed that the fluctuation in the signal
amplitude was quite high. This indicated the onset of tool failure (breakage occurred
Vibration Signature Analysis 703
at hole number 217). The vibration amplitude fluctuation and increase in the frequency
range was observed in all the tests performed to investigate the effect of drill wear.
Several drilling tests were performed under different cutting conditions and it was
observed that the changes in the frequency spectra were not affected by the variations
in the cutting speeds or feeds.
The power spectrum in the Y direction for hole number 217 has a peak at a frequency
of 5.2 kHz. This frequency exists between the modes excited by the chisel and margin
wear. Since this peak frequency does not correspond to any other natural frequencies
of the sub-systems involved (presented in Table 1), it must therefore correspond to a
natural frequency of the combined driU-workpiece system. On the other hand, several
peaks of small amplitudes existed in the power spectrum of vibration measured in the
Z direction. The frequencies of two of these peaks were similar to those excited by
chisel wear. The peak at the 3.3 kHz frequency corresponds to one excited by the
margin wear in the artificial wear test [see Fig. 8(c)].
Figure 14 shows average wear curves plotted for each of the four wear types measured
during the actual drilling tests. The cutting conditions used were the same as those in
Fig. 13. As ,:an be seen from these wear curves, accelerated wear occurred only on
the chisel edge and the margin of the drill.
The data presented in Figs 13 and 14 show that it may be, and indeed it was, possible
to correlate the drill wear to the resulting power spectrum. However, there are several
important difficulties which should be considered.
(1) Materials such as cast iron are not homogeneous and will affect the amplitude
of the vibration measured, and this may cause false alarms.
(2) As mentioned earlier, tool damage in drilling produces a high level of transient
vibrations (spikes) which are largely attenuated by the averaging procedure typically
used in spectrum calculation, and this makes it difficult to extract a discriminating
feature to distinguish the change in the tool conditions.
(3) Non-uniform hardness of the workpiece material, built up edges, and micro-
cracks can also cause false alarms by increasing the vibration amplitude.
In order to minimize the effects of these difficulties, monitoring the change in drill
condition was performed by averaging the power spectra of the vibration measured
during the drilling of M holes, over which wear development can be considered
negligible [20]. The average power spectrum can be calculated as follows:
1,4
,~ 1.2 0.8
1
o.6
o.8
0.6 ~=
.= 0.4
"~ 0.4
a 0.2
0.2
0 o
0 50 100 150 200 250 50 100 150 200 250
Number of Holes Number of Holes
b)
0.3 0.35 j
o.3 .................... . .................... i ............... ............
Fig. 14. Types of wear measured during actual drilling test. V = 12 m m i n - l , f = 100 mm min -1, D = 6 mm,
and L = 3 0 m m .
704 T.I. El-Wardany et al.
1
Fav= ~ , F x x ( f ) (4)
where M represents the number of holes produced with negligible wear development.
In this group of tests it was chosen to be 6 at the start of drilling up to hole number
42, then it increased to 10 for holes numbered 42-151. The value of M was then
reduced to 6 until the end of the tool life.
To avoid the effects of random variations in peak amplitudes, the area under the
vibration spectrum was considered as a monitoring feature for drill wear (rather than
using the variation of the peak at any one frequency). The area under each power
spectrum between two different frequencies, namely, fl and f~, was calculated using
the trapezoid method [20];
A = ~-~(Power (fl) + 2 Power (f2) + ... + 2 Power (J~-l) + Power ~l)) (5)
AMVi
R A M V i - AMVb (6)
where AMVb represents a base line instantaneous Absolute Mean Value calculated at
the start of the drilling process, and AMVi is the current Absolute Mean Value
calculated at the ith revolution of the spindle.
1 i=N
AMViVb = ~ ~ Ixil (7)
i=1
where x i is the instantaneous amplitude of the vibration signal, and the subscript i V b
means i or b. The ratio RAMVi can be used as a setting index for starting the calculation
of the kurtosis or the cepstra ratio. This index or threshold value can be obtained from
a set of calibration tests, or it can be user-defined based on experience.
4. EXPERIMENTALSET UP
Figure 15 shows the experimental set up used in this study. The drilling tests were
performed using a YAM, 2½-axis, CNC machining center. The tools used were 3 and
6 mm diameter high speed steel twist drills and the workpieces were "as-cast" ferritic
nodular cast iron blocks of 290 x 130 x 41 ram, A series of holes were drilled at feeds
in the range of 0.12-0.37 mm/rev, cutting speeds at 12 and 31 m/min, and depths in
the range of 15 to 30 mm. All experiments were performed under dry cutting conditions.
Cast iron was chosen as the workpiece material in order to study the effect of any
Vibration Signature Analysis 705
Eric
2000................................................... o 2000....................................................
Iii i i il
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
HoleNumber HoleNumber
"~~Iiiii2000
1000.....................................................
.....................................................
ii! i i i i !! !i!ii~ iq
HoleNumber HoleNumber
c) V=20 m/min, f=375 mm/min, L =20 mm d) V=30 m/mio, f--400 ram/rain, L =20 mm
Fig. 16. Calculated kurtosis values (K) for different drilling conditions using 3 m m d i a m e t e r drill.
t'
"' .... i
....... r -~r" -1
.!
F I- "
-2
1,1oo I~oo
'nmI~
).~ ......... ~ ........... i "~ ,. i......... ~... ~" .i,I ! ; ':..........i........... i ...... ~ i"-..i ..... ~""i
~... .....~.....-:": • i ~ " - : , . !""i.. i o .,
-I' "~,"~J~ i::"~".'""i":"~'~[:" '"[[?:g'"'"-L :;i ........ .'=:L,,[i, "
:LII:II:i.... !
Fig. 17. Vibration amplitude, power spectra, and cepstra ratio for normal drillin~ process, hole no. 3, drill
diameter was 3 m m (results are shown for R A M V t value > 5 only).
Vibration Signature Analysis 707
ated when setting the RAMVi to 5 for both Z and Y direction. The results represented
in Fig. 16 show that the kurtosis value (K) is insensitive to cutting conditions, or
changes in the: workpiece hardness.
-1
.1 I r',,m~ I r'u,..e
.~ ~i~ wqlvandow:Zlwlndow31window4|
600 900 1~ "26oo m,
i) Y Direction ii) Z d i r e c t i o n
a) VibrationSignals
..... ii •- "., , ....... • -i ¸ ....~
..... i
T ~ ©w ~ o , ~ I 0 OOe~-'
Fig. 18. Vibration amplitude, power spectra, and cepstra ratio generated by a worn drill (corner wear
0.1 ram), hole no. 4 D = 3 mm (results are shown for RAMVi value -> 5 only).
HTH 36-6-D
708 T . I . El-Wardany et al.
-1
-I
°~
35O S~ 75O
• 10 ....... !. :: i • i .: ....
~,,~,,~°" 3 ~ ' . . ~
higher than 5. Figure 17(a) shows the vibration signals (in both Y and Z directions)
used in calculating the cepstra ratio. Figure 17(b) represents the corresponding power
spectra calculated from the vibration signals shown in Fig. 17(a). Figure 17(c) shows
the cepstrum ratio calculated from the vibration signals in both Y and Z directions.
The same procedure was repeated when measuring the vibration signals during the
drilling of hole number 4 with a worn tool (average corner wear was 0.1 mm), which
is illustrated in Fig. 18. In this case, the R A M V i allowed calculation of the cepstra
ratio to be performed for four time windows only, as can be seen in Fig. 18(a).
Although the time needed to drill one hole was 1.2 sec, the calculation of the cepstra
ratio was started after drilling for 0.72 sec as shown in Fig. 18(a). Figure 18(c) shows
that a small peak at a Quefrency of 20 msec (corresponding to one spindle revolution)
appears only once in the cepstra ratio calculated from the vibration signal captured in
time window number 2. Figure 19 represents the case when the tool was broken during
drilling hole number 6. Again, the RAMVi allowed the calculation of the cepstra ratio
to be performed during the drilling time of 0.31-0.715 seconds only. As can be seen
Vibration Signature Analysis 709
0 4000 . . . . . . . .
m breakage
tu %t._ , , I . , l._ .l , . . . . . t
0 5 10 15 20 25 30 35 40 45 50
Quefreney (msee)
(Hole No. 3, V=20 m/min, Time of one revolution=30 msec, f= 375 mm/min, L --20 mm, D =3 mm)
0 2 x 105
t'" d'"
After
"~- breakage
< 1
n" event
t-
oO
n
I
"u°;
0 '
5 'o
1 ;
1 20 ' '
25 3 'o 3'5 40 ;5 fi0
Quefreney (msec)
(Hole No. 5, V=30 m/min, Time of one revolution=20 msec, f= 1100 mm/min, L =15 mm, D =3 mm)
Fig. 20. Cepstra ratio at the breakage event for different cutting conditions.
in Fig. 19(a), three time windows of the vibration signals were processed instead of
the expected 29 time windows. The number 29 was determined from dividing the total
number of points sampled during the drilling of one hole (30,000) by the number of
points (time window) used for calculating the power spectrum and the cepstra, which
is 1024. A peak of large magnitude, at the Quefrency related to the time of one spindle
revolution, appears clearly as a result of drill breakage [see Fig. 19(c)]. Figure 20
shows sample results for the cepstra ratio obtained at different cutting conditions. It
can be seen that the proposed technique is effective for various cutting conditions.
Monitoring of drill breakage using the cepstra ratio is based on the sudden increase
in its value at 1thetime corresponding to one spindle revolution. Therefore, a rectangular
window whose width is about five times the duration of one spindle revolution was
monitored. This proved appropriate for detecting tool breakage. The signals monitored
during drill breakage were affected mainly by the workpiece-spindle resonance, hence,
analyzing the vibration signals up to 5 kHz gave satisfactory results.
5.3. Monitoring of drill wear using the area under the average power spectrum
An R A M V i value of 5 was used to capture the vibration signals in both Y and Z
directions. The number of points used for calculating the power spectrum was 1024,
and 10 data segments were used to perform the averaging process. The power spectrum
was then calculated for each hole. Depending on the number of holes drilled (i.e.
amount of drill wear), the value M was set, and the average power spectrum was then
calculated.
Figure 21 slhows variation of the area under the averaged power spectrum in the Y
direction as a function of the number of drilled holes. As can be seen, these results
showed a trend similar to those observed in Fig. 14.
6. CONCLUSIONS
The development of practical and reliable methods for detecting tool wear and failure
in machining is essential for the realization of intelligent and flexible manufacturing
systems, In this study, the problem of on-line detection of tool failure in drilling
has been studied. Vibration monitoring methods have been developed and verified
710 T . I . El-Wardany et al.
20
15
10
-X--
.l(..
¢1
* -X--
<
- -)(- *
* -1(-
I I I I
50 100 150 200
Fig. 21. Effect of tool wear on the area under the power spectrum curves.
experimentally. The focus of the work was on drills of 6 and 3 mm diameters, used
in conjunction with cast iron workpieces. The main conclusions of this paper are as
follows:
(1) Monitoring vibrations during drilling operation proved useful in predicting drill
wear and failure.
(2) The development of the instantaneous Ratio of the Absolute Mean Value
(RAMVi) was useful in eliminating false alarms, and speeding up the calculations.
(3) Two parameters were selected to detect changes in the pattern of the vibration
signals in the time domain during drill breakage, namely the instantaneous Ratio of
the Absolute Mean Value (RAMVi) and the kurtosis (K).
(4) In the frequency domain, the ratio of the calculated cepstra in the transverse
and thrust directions was used as a monitoring index for detecting drill breakage. The
instantaneous Ratio of the Absolute Mean Value ( R A M V i ) w a s again used to start
the cepstrum calculations.
(5) Monitoring experiments using the small diameter twist drills (3 mm diameter)
verified the effectiveness of both the time and frequency domain methods. Using the
proposed approach for monitoring breakage, it was possible to eliminate false alarms
due to the effect of a sharp drill wandering at the start of the cut. In addition,
experimental evidence showed that changes in the hardness of the workpiece material
had no effect on the proposed monitoring techniques.
(6) The area under the power spectrum curve was used to monitor the wear of large
size drills (6 mm diameter). The RAMVi was used to trigger the calculation of the
Vibration Signature Analysis 711
vibration powe:r spectrum generated during the drilling of a specified number of holes,
where the change in the wear state was minimum. A good correlation was found
between the power spectra calculated and the tool wear.
REFERENCES
[1] A. Thangaraj and P. K. Wright, Machining, Robotics Computer-Integrated Mfg. 4(3/4), 429-435 (1988).
[2] T. Moore and Z. Rief, SME Paper # M584-908, Conference in Sensor Technology for Untended
Manufacturing, Schaumburg, Chicago, Illinois, U.S.A. (1984).
[3] J. Rotberg, E. Lenz and S. Brawn, Mfg. Rev. 3(1), 60-65 (1990).
[4] A. D. Dornfeld and M. F. DeVries, Drill wear and fracture monitoring using acoustic emission,
Presented at CIRP STC "C" meeting, Paris (1992).
[5] J. Tlusty and G. C. Andrews, Ann. CIRP 32(12) 563-572 (1983).
[6] D. Matsushima, P. Bertok and T. Sata, Proc. ASME Winter Annual Meeting, Anaheim, California 55,
145-153 (1992).
[7] E. Brinksmeire, Ann. CIRP 39, 98-101 (1990).
[8] Y. Saijo and A. Noh, A method of detection of drill breakage in drilling, Proc. Autumn Annual
Meeting of JSPE, pp. 11-12 (1978).
[9] T. Hoshi, Proc. 3rd Int. Conf. Automatic Supervision, Monitoring and Adaptive Control in Manufactur-
ing, CIRP, Rydzyna, Poland, pp. 41-58 (1990).
[10] G. S. Li, W. S. Lau and Y. Z. Zhang, Int. J. Mach. Tools Manufact. 32(6), 855-867 (1992).
[11] K. Subramanian and N. H. Cook, ASME J. Engng Ind. 103, 295-301 (1977).
[12] A. Thangaraj and P. K. Wright, ASME J. Engng Ind. U0, 192-200 (1988).
[13] T. I. Liu and S. M. Wu, ASME J. Engng Ind. 112, 299-302 (1990).
[14] P. G. Li and S. M. Wu, ASME J. Engng Ind. 110, 297-300 (1988).
[15] I. N. Tansel, C. Mekdeci and O. Rodriguez, Int. J. Mach. Tools Manufact. 33(4), 559-575 (1993).
[16] T. Moore and Z. Reif, Proc. 13th Int. Conf. NAMRC, pp. 45-50 (1985).
[17] W. K. Yee and D. Blomquist, SME Paper # MR82-901, Dearborn, Michigan Society of Manufacturing
Engineers (1982).
[18] K. W. Yee, SME Paper # MS84-914, Dearborn, Michigan Society of Manufacturing Engineers (1984).
[19] M. Kanai and Y. Kanda, Ann. CIRP 27, 61-66 (1978).
[20] A. Noori-Khajavi, Ph.D. Dissertation, University of Oklahoma State, U.S.A. (1992).