by
M.-H.
Yang and
C.-P.
Chou
A LAB-SCALE DIGITAL ACOUSTIC EMISSION SYSTEM
FOR SOURCE LOCATION
coustic emission
(AE)
techniques have been used *of arrival-time differences (delta-T’s) for each mesh point. To in the study of crack location, cracking mecha- ‘ calculate the source position for a set of burst-type
AE
sig- nisms, deformation of metals and leak detection.:
nals, a straightforward “shift and a d d method was used.A
In these techniques, AE sensors hit by transient.
First, assuming the AE source is at a mesh point inside the stress waves generateAE
signals, and then theAE
signals.
circle, the signal waveform of each channel was shifted for- are analyzed with various methods. Conventional AE sys-.
ward in the time domain, with the corresponding delta-T tems only record some parameters of theAE
signals, such * value. Each time-shifted waveform was added to that of theas peak amplitude, ringdown counts, duration, risetime, and * first-hit channel, and then the resultant additive waveform
so on. On the other hand, a fully digital
AE
system can dig- ’ was squared to give the signal power from that mesh point.itize the waveforms of
AE
signals and perform various anal- ’ When these operations have been performed for all mesh yses. Since many versatile and relatively cheap personal:
points, a power map can be created, and the peak of power computers (PC), analog-to-digital (A/D) cards, and digital.
map is considered the closest point to the AE source. Figures signal processing (DSP) software packages for PC are avail-.
2 shows a set of simulated signals, with theAE
from (1, 21, able, a fully digital and PC-based lab-scaleAE
system for 3-.
propagating at 1480 m/s, of 10 kHz carrier frequency andD source location can be set up. With the results of DSP for
.
sampled at 100 kHz. With this sampling rate, a series of simulated and realAE
signals, features of thisAE
system.
location results for various carrier frequencies of signal (10, and critical factors involved are presented..
20, and 50 kHz) are shown in Figure 3-5. It was found thatwhen the sampling rate was
CONFIGURATION OF 5 times the carrier frequency
AE SYSTEM or higher, source location re-
sults were good no matter where the postulated source The configuration of our
AE
was.
As
the sampling rate system is shown in Fig. 1.lowered, other peaks began Appropriate
AE
sensors andto emerge on the power map. amplifiers can be employed,
When the sampling rate was depending on the purpose of
2 times the carrier frequency, testing. Low-frequency (30
marginally meeting the Ny- kHz)
AE
sensors were usedquist criterion, the location in our leak testing. The A/D
result was incorrect. The cards were two Nicolet
per channel for a total number of 8 channels. This means an
:
approximate 7-times sampling of signals and a time resolu-.
tion of 5 microseconds. Digitized
AE
signals were stored in.
If a signal appears with infinite duration and excellent rep- an Ordinary WINTEL ”, and IUTLAB was wed for data.
etition, as those examples often shown in textbooks when software, so this system was highly flexible.MAIN AMPLIFIER
Fig. I: The configuration of AE system
BE493 A/D cards, working at a sampling rate of 200 kHz ’ Same result was obtained when evaluating with cross-
method.
processing. Not only the
AE
SenSorS and amplifiers were ex- ’ the Nyquist criterion is stated, a sampling rate of two timeschangeable, but also the PC, AID cards and the analysis ’ the carrier frequency is sufficient. In AF, testing or our sim-
SAMPLING RATE
For a digital AE system to perform properly, sampling at a rate high enough is a prerequisite. Because of the nature of irreproducibility of
AE
phenomena,AE
signals simulated by software were used to evaluate the influence of sampling rate. Assume a circle with a diameter of 12 meters is on theXY plane and centered at the origin.
Six
AE
sensors were assumed to have been mounted, equally spaced along the circumference of this circle. Then, an imaginary square mesh with an edge of 12 meters, 0.5-meter interval, also centered at the origin, is placed on the XY plane. This results in 25 X 25 mesh points. Since the propagation speed ofAE
burst is known, the time needed for anAE
burst from any mesh point to reach each sensor is known. That is, there is a setM.-H. Yang is a Research Scientist at Materials Research Lab, Hsinehu, Taiwan. C.-P Chou is a professor in the Department of Mechanical Engineeringat National
cn 1 CH 2 cn 3 cn 4 cn 5 CH 6 (1111” 0 200 400 600 800 to00 Data polntr Chiao-7Lng Llniuersity.
32
EXPERIMENTAL TECHNIQUES JulylAugust 1999x lo‘
Power
Carrier Freq. = 10 Iblr
Sampling Rate = 100 kt-k Peak at ( 1 , 2 ) 1 5 1 0 5 0 6 I
Fig. 3: Source location result when carrier frequency is 10 kHz ‘ ~~ ~ Carrier Freq. = 20 kHz Sampling Rate = 100 W 10‘ 3, Power
5i
Peak at ( 1, 2 ) 2 1 5 1 n s n 6 JFig. 4: Source location result when carrier frequency is 20 kHz Carrier Freq. = 50 kHz Sampling Rate = 100 kHz x lo‘ 2 5 2 1 5 1 0 5 0 6 Peak at ( -0.5, 1 ) Power 6 -6 -6
Fig. 5: Source location result when carrier frequency is 50
kHz
ulation where signals are of finite duration and consist of
noise, a sampling rate of Nyquist frequency seems ins&% cient. As a rough approximation, an AE signal h(t) can be simulated by a sinusoidal function with an exponential de- cay envelope along with noise, that is, h(t) = Ae-“’cos(2.rrft)
+
noise, where A and a are constants, t is time, and f i s the carrier frequency. The frequency spectrum of this signal ex- hibits a bell-shape (due to the exponential decay), and is shifted and centered at f (due to modulation). As the signal becomes shorter (greater a), plus the effect of noise, the bell- shaped curve spreads. This means more frequency (and time) information will be lost if the sampling rate is fked at Nyquist frequency, then the result of data processing will tend to be incorrect.DIGITAL SIGNAL PROCESSING
To pinpoint the position of an
AE
source, the propagation speed ofAE
burst, the hit-sequence, and the delta-T’s must be known. Conventionally, the “threshold-crossing method” is used: when theAE
signal crosses a preset threshold of a channel, this channel is considered “hit” by theAE
burst, and the time of crossing is defined as the arrival time. Then the hit-sequence and delta-”% are deduced thereafter. Threshold setting is user-dependent and somewhat subjec- tive, but usually it is set at 6 dB (two times) higher than background noise. Because of the random nature of noise, in practice the threshold-crossing method is susceptible tonoise, therefore leading to erroneous location results.
To make our system more robust to noise, a series of digital signal processing techniques were used. First, a digital band- pass (BPI filtering was applied to each digitized waveform of
AE signal, according to the nominal frequency of AE sensors used. For instance, a BP filter of 20-40
k H z
is used for AEsensors of 30 kHz. Second, filtered
AE
signals were trans- formed to their corresponding analytic signals’. With an an- alytic signal, the information of amplitude vs. time is kept but the oscillation term is suppressed. Third, the threshold- crossing method was applied to determine arrivals. Finally, the cross-correlation between two analytic signals was per- formed to give the relative arrival-time between two chan- nels, then the hit-sequence and delta-T’s were determined. Since the maximal value of cross-correlation indicates the relative time-shift between two main energy packets, correct hit-sequence and delta-T’s can be deduced even in the pres- ence of noises. Figure 6 shows a set ofAE
signals due to the leakage through a drilled hole in a leak experiment.As
shown in Figure 6, in source location calculation, it is a com- mon practice to set a certain time window which at least is equal to the maximal possible arrival-time difference be- tweenAE
sensors, and then analyze the signals within the time window only. This also reduces calculation time signif- icantly. The corresponding normalized analytic signals within the time window are shown in Figure7.
The noise was rather low, so the main energy packets (designated by“P)
can be easily identified. Figure 8 shows another set ofAE
signals that are weaker and severely “contaminated by noise. Since there are many “noise packets” (designated by “n” in Figure 91, conventional threshold-crossing methodwill
lead to incorrect result. However, by using cross-correlation the correct result still can be obtained.
I I 0 cilculilcd AE source X sensor + drlUed hole/ / 0 2 - /
1
~ 0 1 - 0 - (melera) - 0 1 - 0 -0 2 - \ / /' - 0 3 - 0 2 -01 0 0 1 0 2 0 3Fig. 6 A set of AE signals obtained in a leak experiment.
The partial loss of couplant resulted in the lower signal amplitude of channel 6.
I
cH4L
PP
f l \
1
C H0 6 h20 _ _40 _ _ .60 . - 880 100 120 140 160 ' ' 180data points ( 1 point = 5 microseconds)
Fig. 7: The normalized analytic signals of AE signals within
the time window in Figure 6. The main energy packets (C'P")
can be easily identified.
1 0 ma 0 1 1 N I . .I I am* a l l 1 1.1 1 mma a, " I 1.1 1 mom U 5 " * 11. 1 **a Ma " t (*a t 57.
Fig. 8: Another set of AE signals obtained in a leak
experiment
P
n n /-,\,
CH 6 _i - 8 - 8 -
0 20 40 60 80 100 120 140 160 180 data points ( 1 point = 5 microseconds)
I
'
Fig. 9: The normalized analytic signals of AE signals within
, the time window in Figure 8. Many noises (C'n") are present.
Fig. 10 Some location results of the AE signals in the leak
experiment
:
3-D SOURCE LOCATION.
When hit-sequence and delta-T data are available, calcula-'
tions can be conducted to determine the location of an
AE
'
source. Various algorithms of AE source calculation have be-
:
ing developed, but the Spherical Interpolation 61) method.
for 3-D location by Smith and Abe12 was adopted. The SI.
method consists of a series of matrix computations, and is.
implemented with MATLAB. Some location results of theAE
.
signals from leakage are shown in Figure 10, which is the.
projection of 3-D locations on the =-plane. The calculated * positions show good consistency and are very close to the' leak hole.
q CONCLUSIONS
:
A PC-based, fully digital and lab-scale acoustic emission sys-.
tem for 3-D source location was set up. To get good results,.
a sampling rate higher than the Nyquist frequency is pre-ferred. and is a t least five times the carrier freauencv of sia- ‘ References
nal for our experiment. With digital filtering, aialytical sig- 1, Gammell, P.M., “Improved Ultrasonic Detection using the
nals, and cross-correlation technique, the source location op-
.
Analytic Signal Amplitude, 7) Ultrasonics, 73-76 (Mar. 1981). eration can be made more robust to noise. Versatile 3-D.
source location calculation can be implemented with numer-
.
2. Smith, J.O., Abel, J.S., “Closed-Form Least-Squares Source.
Location Estimation from Range-Difference Measurements, ” IEEEical and DSP software packages.
.
Transactions on Acoustics, Speech, and Signal Processing, ASSP-35.
(121, 1661-1669 (Dec. 1987).ACKNOWLEDGMENT
This work was supported by Chinese Petroleum Corporation,
.
Taipei, Taiwan.
I I