U.S. patent application number 10/861369 was filed with the patent office on 2005-12-08 for method and system for non-destructive evaluation of conducting structures.
Invention is credited to Doorn, Eric van, Haynes, Leonard S..
Application Number | 20050270037 10/861369 |
Document ID | / |
Family ID | 35446980 |
Filed Date | 2005-12-08 |
United States Patent
Application |
20050270037 |
Kind Code |
A1 |
Haynes, Leonard S. ; et
al. |
December 8, 2005 |
Method and system for non-destructive evaluation of conducting
structures
Abstract
Method and system for non-destructive evaluation for a
conducting structure by measuring the electrical impulse response
thereof including applying a PRBS test input signal to the
conducting structure, detecting an output signal from the
conducting structure and processing the data to assess the
condition of the conducting structure via changes in the electrical
impulse response and to locate any defects along the conducting
structure.
Inventors: |
Haynes, Leonard S.;
(Rockville, MD) ; Doorn, Eric van; (Frederick,
MD) |
Correspondence
Address: |
EPSTEIN & GERKEN
1901 RESEARCH BOULEVARD
SUITE 340
ROCKVILLE
MD
20850
US
|
Family ID: |
35446980 |
Appl. No.: |
10/861369 |
Filed: |
June 7, 2004 |
Current U.S.
Class: |
324/603 |
Current CPC
Class: |
G01R 27/28 20130101;
G01R 31/088 20130101 |
Class at
Publication: |
324/603 |
International
Class: |
G01R 027/02 |
Claims
What is claimed is:
1. A system for non-destructive evaluation of a conducting
structure comprising generator means supplying a series of high
frequency input signals to the conducting structure; means
providing a time delayed replica of each of said input signals; an
analog multiplier receiving an output signal from the conducting
structure in response to each of said input signals, receiving said
time delayed replica of each of said input signals and producing a
multiplication output thereof; a sample and hold circuit receiving
said multiplication output and supplying unsynchronized samples
forming an impulse response function output; an analog-to-digital
converter receiving said sample and hold circuit output; and a data
processor receiving an input from said analog-to-digital converter
to produce an average signal from a fixed number of samples, to
initiate successive series of input signals and time delayed
replicas of each input signal with increasing time delays and to
produce an average signal from each successive time delay of said
time delayed replicas of said input signals creating, in aggregate,
the impulse function of the conducting structure.
2. The system for non-destructive evaluation of a conducting
structure recited in claim 1 wherein said data processor means
includes means for performing principle component analysis of said
impulse response.
3. The system for non-destructive evaluation of a conducting
structure recited in claim 2 wherein said input signals are PRBS
signals.
4. The system for non-destructive evaluation of a conducting
structure recited in claim 3 wherein said PRBS signals have a
frequency greater than 20 MHz.
5. The system for non-destructive evaluation of a conducting
structure recited in claim 1 and further comprising an input
inductive device for coupling said input signals to said conducting
structure.
6. The system for non-destructive evaluation of a conducting
structure recited in claim 5 and further comprising an output
device disposed adjacent said input inductive device for detecting
said output signal from said conducting structure.
7. The system for non-destructive evaluation of a conducting
structure recited in claim 1 wherein said generator means includes
shift register means supplying said input signals and said time
delayed replicas of said input signals.
8. A method for non-destructive evaluation of a conducting
structure comprising the steps of supplying a series of high
frequency input signals to the conducting structure; deriving a
series of output signals from the conducting structure, each output
signal corresponding to one of the input signals; generating a time
delayed replica of each of the series of high frequency input
signals; multiplying the time delayed replica input signals and the
output signals to produce a multiplication signal; sampling the
multiplication signal at unsynchronized discrete points in time to
produce an impulse response function; converting the impulse
response function to a digital signal; and processing the digital
signal to produce average signals from at least first and second
time delayed series of generated signals.
9. The method for non-destructive evaluation of a conducting
structure recited in claim 8 wherein said processing step includes
principle component analysis, a clustering algorithm and a final
diagnosis algorithm.
10. The method for non-destructive evaluation of a conducting
structure recited in claim 8 wherein said supplying step includes
supplying PRBS input signals.
11. The method for non-destructive evaluation of a conducting
structure recited in claim 10 wherein said supplying step includes
supplying PRBS signals having a frequency greater than 20 MHz.
12. The method for non-destructive evaluation of a conducting
structure recited in claim 10 wherein said supplying step includes
supplying PRBS signals having a frequency range between a few to
hundreds of MHz.
Description
BACKGROUND OF THE INVENTION
[0001] The present invention pertains to methods and systems for
evaluating the structural integrity of electrically conductive
structures and, more particularly, to such methods and systems
wherein the electrical impulse response of the structure is
measured after a test input signal is applied in order to assess
the integrity of the structure and the presence, severity and
location of any defects.
[0002] The prior art, as exemplified by U.S. Pat. No. 3,988,667 to
Roth et al, U.S. Pat. No. 4,067,060 to Poussart et al, U.S. Pat.
No. 4,275,446 to Blaess, U.S. Pat. No. 4,935,699 to Boenning, U.S.
Pat. No. 4,988,949 to Boenning et al, U.S. Pat. No. 5,025,221 to
Blaess, U.S. Pat. No. 6,064,212 to Ariweilar et al and U.S. Pat.
No. 6,265,880 to Bomet et al and British Patent No. 1,160,271 to
Booth et al, discloses the use of periodic-random, noise and
pseudo-random binary sequence test input signals for measuring
transfer functions of systems under test, and/or measuring
transmission characteristics by Fourier analysis, and/or systems
for detecting chafing of cables and/or conduits; however, the prior
art does not permit implementation of high frequency evaluation for
detecting subtle changes of electrical characteristics of
structures to be tested or evaluated. In prior art methods for
testing of servo systems, the impulse response function is compared
with a set of known correct functions and any discrepancy is
noted.
[0003] It is known that the cross correlation of a random noise
input and the resulting output of a system is identical to the
impulse response of the system. To be perfect, the random noise
input would need to be infinitely long and perfectly random. In
practice, a specific length, pseudo-random binary signal has been
disclosed, as for example in the Booth et al British Patent No.
1,160,271, where the cross correlation function of the input with
the output is generated by computing the average of the cross
correlation of the output with a copy of the input delayed to
generate one point on the impulse response function. An integrator
is used to measure the average value of the impulse response; and,
accordingly, the Booth et al system has the disadvantage of not
being feasibly implemented for high frequency test signals.
SUMMARY OF THE INVENTION
[0004] The present invention overcomes the disadvantages of the
prior art in non-destructive evaluation of a conducting structure
by measuring the electrical impulse response thereof. The term
"conducting structure" as used herein means any object or system
having an electrical impulse response upon application of an
electrical test signal thereto and includes, but is not limited to,
cables, wires, including insulation around wires, fencing, circuits
and other objects having electrical characteristics exhibiting
changes, no matter how subtle, that can be detected in accordance
with the present invention. Accordingly, "conducting structure" as
used herein includes partially conducting structures of the type
where conductivity occurs internally and/or externally of the
structure. Such conducting structures may have a very high
resistance relative to DC voltage such as to be essentially
non-conductive to DC.
[0005] Another object of the present invention evaluates a
conducting structures by applying a test signal to a convenient
input point of the conducting structure and measuring an electrical
impulse response at a convenient output point of the conducting
structure. The output and input points can be nearly co-located
facilitating use with elongated conducting structures.
[0006] In accordance with another aspect, the present invention
uses a test signal of a frequency sufficiently high that the
wavelength of the test signal and, thus, the skin depth of the
propagating energy, is small and, further, uses a very wideband
high frequency signal so that there is great spectral richness in
the test signal.
[0007] The present invention uses an analog multiplier to multiply
a time-shifted, pseudo-random binary sequence (PRBS) input signal
with an output signal from a conducting structure and samples the
resulting waveshape with sampling instants asynchronous with
repetition of the PRBS input signal and asynchronous with the
multiplication of the time shifted input signal and the output
signal.
[0008] The present invention tests conducting structures using test
signals having a very large bandwidth at a high frequency and,
consequently, a very small skin depth such that energy travels near
the surface of the conducting structures or in the material around,
including air, just outside the conducting structures, the large
bandwidth creating great spectral richness in the test signal.
Changes in the conducting structures near the surface cause complex
patterns of multipath reflections in the conducting structures and
are detectable as changes in the electrical impulse responses of
the conducting structures. In this manner, the present invention
can detect changes in conducting structures manifested by cracks or
abrasions in wire insulation, cracks, corrosion or stress in piping
or other elongated structures, such as fencing or wires or metal
bands, or fluid on the surface of the conducting structures.
Similarly, the present invention can be used to measure the ionic
content, such as chlorine, of fluids.
[0009] An additional aspect of the present invention is to permit
testing of elongated conducting structures, such as cables, wires,
piping and structural members, with input and output points at the
same end of the conducting structure thereby providing ease of
installation and testing and reducing the output signals required
for analysis as compared with the case where the input and output
points are at opposite ends of the conducting structure.
[0010] Another aspect of the present invention is the use of a
pseudo-random binary sequence (PRBS) where all binary values with M
bits are used once and only once for an M length sequence, such a
code being capable of being generated using a properly connected
shift register generator. The clocking frequency of the PRBS is a
high frequency, near 70 MHz in a particular embodiment; and, an
analog multiplier multiplies the time shifted input signal with the
output signal. The resulting waveshape is sampled with the sampling
instants being asynchronous with the repetition of the PRBS and the
time shifted input and output. With a 70 MHz clocking frequency,
the resulting PRBS will have significant spectral content from a
few MHz to several hundred MHz. The time delays (shifts) need not
be smaller than the clock cycle time of PRBS for detection of
defects but should be smaller than the clock cycle for the precise
localization of defects.
[0011] Some of the advantages of the present invention over the
prior art include use of the present invention for testing any
structure which is conducting to any extent at high frequencies
such as, for example, wiring, piping, fencing and circuitry, use of
the present invention for testing chlorine level and pH level in
water and other fluids, use of the present invention for testing
for fluid leaks, such as of water and hydraulic fluids, use of test
signal frequencies of a few MHz to hundreds of MHz, requiring
simple, low cost hardware. Because of the high frequency of the
test signals, the test signals can easily be separated from normal
power frequencies (such as 60 MHz or 400 Hz) by using a simple
passive filter thereby enabling built-in, on-line testing of
electrical power wiring.
[0012] Each time delay t is repeated many times, and one or more
samples from each repetition (asynchronous) are captured and
stored. The average of the samples is then used as the impulse
response function for the point t on the function. Generally, a
fixed number of samples would be used such that statistically the
correct average value is achieved. Because the frequency is very
high, many samples can be used for each point on the impulse
response function and still generate the complete impulse response
function in a few milliseconds. Very subtle changes are detected in
the impulse response function due to changes in multipath
reflections and interference that occur over a very wide frequency
range as a result of insulation defects for wire, abrasions on
piping, water or hydraulic fluid on piping, etc. Since direct
comparison with known normal impulse response functions would be
inadequate, typical impulse response functions are processed with
mathematics known as Principal Component Analysis (PCA) to generate
what are known as Principal Components.
[0013] Other aspects and advantages of the present invention will
become apparent from the following description of the preferred
embodiments taken in conjunction with the accompanying drawings,
wherein like parts in each of the several figures are identified by
the same reference characters.
BRIEF DESCRIPTION OF THE DRAWINGS
[0014] FIG. 1 is a block diagram of a system according to the
present invention.
[0015] FIG. 2 illustrates the primary processing functions for
implementing the system of FIG. 1.
[0016] FIG. 3 illustrates the processing functions to produce
identification and quantification for implementing the system of
FIG. 1.
[0017] FIG. 4 shows a typical PRBS signal for use with the present
invention.
[0018] FIG. 5 shows the output from a system receiving the PRBS
signal of FIG. 4 as an input.
[0019] FIG. 6 shows an impulse response function derived from the
waveform of FIG. 5.
[0020] FIGS. 7 and 8 show amplitude and phase functions,
respectively, derived from the waveform of FIG. 5.
[0021] FIG. 9 is a diagram of a modification of the system of FIG.
1.
DESCRIPTION OF THE PREFERRED EMBODIMENT
[0022] It is known that cross-correlation of test input and output
signals supplied to and received from a device under test contain
sufficient information to evaluate the integrity of the device. It
is an important feature of the present invention that the
cross-correlation can be accomplished without acquiring the input
and output signal voltages directly and without fast digitization
of the input and output signals. To this end, in accordance with
the present invention, the test input signal (or replica thereof)
is delayed and supplied to an analog (voltage) multiplier along
with the output signal such that the output of the multiplier
represents the time-averaged correlation function for a particular
input-output leg. The multiplier output can be digitized at a
reduced speed due to the interest in time-averaged signals.
[0023] A preferred embodiment of a system according to the present
invention is shown in FIG. 1 for testing a conducting structure 100
wherein a PRBS test signal is coupled with an input point 110 of
the conducting structure and an output signal is taken from an
output point 120 at the other end of the conducting structure. The
PRBS test signal is generated by a shift register generator 200.
The clock speed required is quite high, and the shift register
generator is generally implemented as a hardware shift register
with feedback from selected bits connected through "exclusive or"
gates in the conventional manner. The exact connection of the bits
of the shift register generator determine the exact sequence of
bits at the output 210 and are selected to yield a perfect code
with 2.sup.n distinct values of the shift register's n bits for a
shift register of length n. A clock signal 300 invokes each
successive shift. A typical frequency for the clock in a preferred
embodiment is 70 megahertz (MHz). The sequence is initialized with
a start signal 700. A delayed shift register generator 500
generates a delayed bit sequence identical to that generated by
shift register generator 200 except delayed in time by d clock
pulses. The amount of delay is determined by a control processor
400 and can be implemented by preloading n bits into the delayed
shift register generator 500 so that d clock pulses are required to
get to the initial value of shift register generator 200. Control
processor 400 selects the delay value from a table stored in its
memory, places the delay value on the direct input port of the
delayed shift register 500 and raises the load signal 510. It then
gates successive clock pulses to the delayed shift register
generator through clock input 520. The output of the delayed shift
register generator 500 is passed to a level shift circuit 530 which
shifts the level of binary signal to provide an output 540 to best
match an analog multiplier 1000. Sub-chip delays (delays smaller
than one clock cycle) can be achieved by using a clock multiplier
with phase control, such as Texas Instruments part CDCF5801.
[0024] The output signal from the conducting structure 100 is
coupled to a normalization circuit 900 which adjusts the level of
the output signal to match the input signal range of the analog
multiplier 1000 and results in the analog multiplier output
generally remaining in its linear range and on average using a
large percentage of the dynamic range of t he analog multiplier.
The design of normalization circuit 900 is simply accomplished
since the other input 540 of the analog multiplier 1000 is a binary
signal and is always either zero or a fixed voltage level. Analog
multiplier 1000 operates at very high speeds and has a bandwidth at
least up to 200 MHz. Even higher frequency ranges may be useful for
other applications. An example of a suitable analog multiplier is
the AD835 marketed by Analog Devices. The output 1010 of the analog
multiplier is supplied to a sample and hold circuit 1100 controlled
by a sample signal clock 2 1020 to initiate the capture and hold of
whatever value was present at the instant of the signal clock 2
1020. Signal clock 2 is asynchronous and uncorrelated with the
shift register generator clocks 300 and 520. Any convenient clock 2
speed can be used; but, the faster the sample and hold circuit
operates, the faster results will be available.
[0025] Sample and hold output 1110 is digitized by an A/D converter
circuit 1200. In the preferred embodiment, a 12 bit conversion is
adequate. The digital output of the A/D converter 1210 is supplied
to the input of a microprocessor 1300 for the remaining processing.
It is important to understand that the A/D conversion and storage
in the microprocessor does not need to occur at high speeds even
though the signals of interest are as high as 200 Megahertz or even
higher in other applications. The first processing step to be
performed by microprocessor 1300 is to compute the average of a
fixed number of samples taken from repetitions of the impulse
response function. Once that number of samples has been taken, the
control microprocessor 400 will initiate the next delay and begin
repeating the pseudo-random sequence again, and the microprocessor
1300 will begin computing the next average value for that next
delay. Alternatively, a varying number of samples can be averaged
until the average converges. This has the potential to require less
samples than if a fixed number of samples are used, but it adds
some processing complexity. For most applications, the speed of
testing is not critical, and using a fixed number of samples is
simpler.
[0026] Instead of directly comparing a captured impulse response
with previously stored impulse responses of a properly operating
system; in accordance with a preferred embodiment of the present
invention impulse response data compressed into a form which
assists in change detection, quantification and diagnosis. As shown
in FIGS. 2 and 3, three values are particularly important. n is the
number of points per impulse response function. Typically, this
would be in the range of 1000 to 5000 points per function. k is the
number of traces used to characterize a given conducting structure.
m is the number of principal components used in classification and
result generation, i.e. n=points per impulse response, k=number of
traces and m=number of principal components. As shown, the output
1210 of A/D converter 1200 supplies data to a primary component
analysis (PCA) processor 1400, which data is stored in an array
S(i,j) i=0,n; j=0,k. S(0,1) S(1,1) S(2,1) S(3,1) . . . S(n,1) is
the data from the first impulse response. Each successive point i
is the result of the next delay as described above. There are n
such delays and hence n points in the data for S(i,1). For PCA
processing, k of these traces are stored in memory, indexed by
index j=1,k. The PCA algorithm yields two primary results. The
first result, 1410, is array A(k,l) k=0,n; I=0,m. Index k is from 0
to n where n is the number of points in each trace. Index I is for
each principal component. A(0,1) A(1,1) A(2,1) A(3,1), . . . ,
A(n,1) defines the first principal component as the weighted sum of
the n points that make up the original data. The algorithm which
computes the principal components selects the weights to maximize
the variance in the data in a manner which can be found in most
textbooks on statistics and signal processing. The first k
principal components are computed. Principal components are
computed when the system is known to be operating correctly, and
then these components are fixed. After that time the values of A
are treated as constants and applied to incoming data to generate
array O(p,q) as described below.
[0027] The second result, 1420, of the PCA algorithm is the array
O(p,q) p=0,m; q=0,k. The matrix O stores the coordinates of the
incoming raw impulse response transformed into the new coordinate
system defined by the matrix A. O is computed by multiplying matrix
S by matrix A as shown in the following equation. 1 S ( i ) i = 0 ,
n * ( Ak )
[0028] As shown in FIG. 3, a classifier 1500 uses the data O (p,q)
expressed in the reduced dimensional space generated by the PCA
algorithm results 1410 and 1420. This is the output data in the
coordinate system defined by the PCA algorithm and the resulting
principal coordinates. It also uses as input any a priori results
R(v,w) v=0,m; w=0,k which classify and/or quantify known abnormal
clusters as to the cause of the abnormality. The a priori results
are used to help define clusters and to help quantify clusters as
they relate to desired values, such as the severity of damage to
the conducting structures being tested. The processing represented
by the clustering algorithm of the classifier 1500 is performed
off-line using nominal data acquired when the system is known good
and, when feasible, data resulting from specific faults. As a
generalization, the more data which can be fed into the clustering
algorithm for training the better. Since this processing is done
infrequently, the time it requires is unimportant. The output of
the clustering algorithm is a partition definition 1530 that
partitions the space defined by the PCA principal components into
areas representing normal and abnormal and also representing
quantification of changes. As examples, in the case of
non-destructive testing of hydraulic lines, specific areas of the
space may represent minor abrasion, and other areas may represent
severe abrasion. The learning or training process can proceed
continually allowing the system to slowly adapt to long term
changes, if desirable. However, in the case of non-destructive
testing, it is generally not desirable to allow the system to
slowly adapt to changes since it is exactly those long term changes
from the original known good system that are to be detected and
diagnosed. A result generator 1600 provides on-line processing that
takes as input the partition definition 1530 and the next real-time
sample impulse response function 1620 represented in the PCA space
and computes the distance of that point from the various nearby
clusters and, from those distances, draws a final conclusion. If
the data is in an abnormal cluster, the result generator identifies
and quantifies the probable cause as partition identification 1630.
If the new data is not in any cluster where a diagnosis is
available, the system merely identifies the data as abnormal as
numerical result 1640.
[0029] There follows a more detailed explanation of the mathematics
and underlying science concepts and the data processing. For the
purpose of ease of description, the system and method of the
present invention will be described primarily in connection with
wires and circuits, particularly detection of chafing of cables;
however, it is understood that the system and method of the present
invention can be used to test any conducting structures as defined
above.
[0030] Principle component analysis (PCA) is a tool that can reduce
signal dimensions so that one can visualize the data clusters very
easily in 2D or 3D dimensional space. The following best
illustrates the key idea of PCA. 1
[0031] The most important property of PCA is its capability of
dimensionality reduction. One may reduce the number of output
features needed for effective data representation by discarding
those linear combinations that have small variances, and retaining
only those terms that have large variances. Let .lambda..sub.1, . .
. , .lambda..sub.p be the largest m eigenvalues of the correlation
matrix R. The data vector x is approximated by truncating the
expansion in the above expression after p terms as follows: 2 x ^ =
j = 1 p a j u j , p < n
[0032] Where n is the dimension of x, a.sub.j, j=1, . . . , p is
the projection value of x to the j-th eigenvector u.sub.j, and
a.sub.1, . . . , a.sub.p are called features of x. In many cases,
p=2 or 3 are chosen so the features can be visualized in 2D or 3D
space. The eigenvalues .lambda..sub.p, . . . , .lambda..sub.n-1 are
the smallest (n-p) eigenvalues of the correlation matrix R; they
correspond to the terms discarded from the expansion of x to
construct the approximating vector x'. The closer these eigenvalues
are to zero, the more effective the dimensionality reduction will
be.
[0033] When wire chafing is being tested, for example, data might
be taken for six different levels of chafing damage. There would be
six sets of test data, one for each level of chafing. The six
levels would be w=0.0, w=0.06, w=0.12, w=0.24, w=0.48, w=0.96
inches, respectively, where w denotes the width of the chafed area.
For each chafing level five hundred series of input and output
signals would be acquired.
[0034] Thereafter, the test data would be averaged within each test
set or chafing level. Because of measurement noise and other
unknown factors, using only one individual measurement signal may
not reveal the real dynamic relationship between the acquired input
and output signals. Therefore, filtering techniques would be used
to mitigate the noise impact as well as the effects caused by other
unknown uncertainties. For each chafing level, five hundred pairs
of input and output signals would be acquired with each signal
containing five hundred data points.
[0035] Using the first input voltage time series as-a reference,
the time lag of subsequent time series is determined by means of a
cross-correlation operation. In this way, all five hundred time
series can be "lined up" for both input and output voltages. After
the signals are acquired and saved to data files, they are evenly
divided into five groups. Each group has one hundred input and one
hundred output signals. For each chafing level, five averaged pairs
of input and output signals result.
[0036] Performing a cross correlation operation on the averaged
data signals, the outputs will be 30 column vectors,
x.sub.i.epsilon.R.sup.999- ,i=1, . . . , 30.
[0037] Let X=[x.sub.1, x.sub.2, . . . , x.sub.30] and R=X.sup.TX .
Find two principal eigenvectors, U.sub.i.epsilon.R.sup.999, i=1,2,
of R, that correspond to the two largest eigenvalues of R.
[0038] Project x.sub.i,i=1, . . . , 30, to the two principal
eigenvectors to obtain a set of features with each feature vector
containing two values.
[0039] Plot the features in 2D space, i.e. feature space. The
result is shown below where different colors would represent
different chafing levels.
[0040] Levels one (normal case, unchafed wire), two and three are
grouped together, and levels four, five and six are also grouped
together. Basically, this reveals that when the chafing level is
below 0.12 inches, the PCA can hardly detect the anomaly. However,
when the chafing level exceeds 0.24 inches, the PCA features are
very distinct from ones with no chafing. Furthermore, a large
chafing amount, for example, 0.96 inches, doesn't push the PCA
features further away from the normal cluster.
[0041] Instead of averaging the raw input and output signal,
averaging can be performed after the correlation. Numerically, this
will increase the computation complexity. However, averaging after
the correlation is, practically, a better approach as there are
many commercial off-the-shelf hardware correlators available, some
with averaging functions.
[0042] For practical applications, the injection (input) and
detection (output) of the PRBS probe signal can take place close
together, so that only one access point is needed to inspect an
entire wire.
[0043] In accordance with the present invention, a small PRBS input
signal is injected into the conducting structure , such as a cable,
and the cable's distortion due to changes in the wire
characteristics is studied. Very low level PRBS signals can be
added to normal cable inputs, and a complete characterization of
the cable can be obtained while the cable is performing its usual
task in that the PRBS signal can be sufficiently low level to not
affect nominal cable operation. The output at the test point is
then averaged through many probe (PRBS input signal) sequences to
obtain the desired impulse response. Since the testing can be fully
built-in and on-line, the time required to complete the averaging
is unimportant. The approach can also be used to test cables
off-line where the cables are not carrying other signals during the
test.
[0044] The system of the present invention can be easily
implemented on a single integrated circuit and can be used to
perform constant on-line prognostic characterization of cables.
Because it can be implemented on a single IC, it is fully feasible
for built-in test applications. Microprocessors are available that
include integral A/D converters, read-only, and writable memory.
With these components, it is quite feasible that only a single
microprocessor is necessary to implement the present invention.
[0045] As noted above, a PRBS signal is applied to the input of any
conducting structure, such as a circuit or cable, as a test-probe,
the PRBS being a waveform of binary pulses which can be of varying
(but precisely-determined) numbers of pulses with different
durations and positions in the sequence. A typical PRBS waveform
consists of 255 pulses applied in about 25 msec and is shown in
FIG. 4. When the PRBS signal is inserted into the input of a
circuit such as an inverting op-amp with a DC gain of -4.7, and a
parallel RC circuit in the feedback loop, the result at the output
is shown in FIG. 5.
[0046] From mathematical manipulation of the waveforms, the impulse
response function of the circuit, which is shown in FIG. 6, can be
derived. By further manipulation, it is then possible to derive the
complex transfer function of the circuit in amplitude-phase form.
This is essentially a Bode-plot in non-logarithmic coordinates,
showing the amplitude of the transfer function at every frequency
in one plot, and the phase lead/lag in the other as shown in FIGS.
7 and 8 respectively.
[0047] These plots completely characterize the input/output
relationships of the circuit under test. Any change whatsoever in
the operation of the circuit which affects its input/output
relationships must be reflected in some change in these plots. As a
result, they are the ideal basis of comparison between the in situ
measured response of a circuit and its known, good, expected
response, which may be stored in ROM for comparison. In the
preferred embodiment of the present instant invention, the raw
impulse response of the known good system is neither stored nor
compared directly. Instead, the data is compressed using the PCA
algorithm as described above, and the position of the data points
in the reduced dimensional space are compared using cluster
analysis.
[0048] As will be appreciated from the above, in accordance with
the present invention, a low level PRBS signal input is injected
into a conducting structure, such as a cable. The correlation of
the input with an output signal received from the cable gives the
impulse response of the circuit. That impulse response completely
characterizes the transfer function of the cable, including the
analog properties of connectors as well as wires. This is a
complete characterization of the transfer function of a cable (and
any other included components), including its frequency and phase
characteristics over a wide frequency range. If the cable changes
due to a high resistance point, leakage, changing capacitance to a
shield or to another conductor in the cable, its transfer function
will change and is detected and diagnosed. Similarly,
discontinuities which cause signal reflections and hence standing
waves in the cable are detected and diagnosed. The method is
independent of either signals or power on the cables, so the method
can be used for on-line all-the-time testing or off-line
testing.
[0049] The method provides a measure of the health of the cable,
giving its actual transfer function. This allows prognostics as
well as diagnostics to be performed. Consider a digital data cable
where digital data may be transferring properly, but the cable is
deteriorating due to carbonization which is reducing the analog
bandwidth of the cable and increasing crosstalk with other wires in
the cable bundle. The testing method of the present invention will
detect this situation and, with appropriate built-in test hardware,
will be able to test continually as the cable performs its normal
function. External testing is also very feasible using the method
of the present invention. Performing an FFT on the impulse response
gives the circuit's complex transfer function, similar to a Bode
plot and, therefore, gives its gain and phase characteristics at
all frequencies of interest. If the cable deteriorates such that
its transfer function changes, the impulse response must change and
can be detected by comparing the nominal transfer function with the
measured transfer function. The FFT is not necessary in accordance
with the present invention. Furthermore, since the complete
transfer function is produced, failures can often be diagnosed (as
well as detected) by observing the details of the change in the
transfer function from its nominal case. The low level PRBS signal
that is used as the "probe" in the present invention can, in
theory, be an arbitrarily small signal, even below the noise level
of circuits on either side of the cable. This is possible because
the resulting signal is averaged over relatively long times to
obtain the result. The lower the signal level, the longer an
averaging time is required but, in principal, the PRBS "probe" can
be made sufficiently small that it cannot interfere with the
operation of the cable, yet the complete transfer function of the
cable can still be obtained. This is particularly effective when
the cable is carrying low frequency signals such as power because
the averaging necessary to separate the PRBS probe or input signal
and the power will be very short, providing near real-time
detection of faults.
[0050] The cross-correlation between the input and output contains
enough information to detect chafing and other defects, and the
cross-correlation can be formed using hardware without the need to
acquire the input and output voltages directly (but only their
products) and without the need for fast digitization of any signal.
If the signal injected into the wire and the output signal are fed
in to a controlled delay and multiplier, the time-average of the
resulting signal will correspond to the value of the time-averaged
correlation function for a particular input-output lag. The signal
from the multiplier can be digitized at a much lower pace than the
raw input and output signal because there is only interest in the
time average. In fact, care should be taken not to sample too fast
in that a correct estimate for the time average will only be
obtained if two consecutive samples are at least one correlation
time apart. For the signals used, the autocorrelation times are
roughly one clock cycle, or {fraction (1/64)} MHz such that
sampling can be at a pace dictated by the digitizing capability of
a low-end microprocessor. For instance, if digitizing at 10K
samples per second, the entire cross-correlation (averaged over
five hundred PRBS-codes) could be constructed in about ten
seconds.
[0051] The low level PRBS signal can be induced on wiring using an
inductive loop or coil, eliminating the need to unplug a cable and
connect directly to the cable to get the test signal onto the
cable. If the coil has been selected properly, the PRBS generator
will not be significantly loaded by the coil, and the delayed
signal and the signal fed into the coil will contain essentially
the same waveforms.
[0052] FIG. 9 shows the system of the present invention where the
input coil 2110 and the output coil 2120 are coupled to the
conducting structure 100 adjacent the same end thereof and also
shows the insertion of a filter 2130 in the output line in the
event the conducting structure is passing operating signals, such
as 60 Hz power, therethrough as described above. While the coils
are shown adjacent the conducting structures, the coils can also
extend around the conducting structure, the primary consideration
being the establishment of magnetic flux coupling.
[0053] Specific parameters need be considered (and the constraints
on them) for optimal design of the coils, both for injecting the
PRBS-code into the cable, and for picking up the signal. The
parameters of the coil are its resistance R, its inductance L, the
number of windings N, and the radius of each winding, r. The coils
do not necessarily need to be looped around the conducting
structure; holding the coils in close proximity to the conducting
structure will provide coupling through magnetic flux.
[0054] For both injecting the PRBS signal and pickup of the output
signal, the first consideration is the response time of the signal:
the coil needs to be fast enough to deal with the 70 MHz clock
cycle (and also with the rise time of the binary data, which is on
the order of 5 ns) in the embodiment discussed herein. The response
time .tau. of a coil is determined by its inductance, L, and its
resistance, R: .tau.=L/R. Thus, the constraint that
.tau.=L/R<<1/70 MHz. Considering the input coil, maximum flux
should be injected into the cable. This means r should be as small
as possible (snug fit around cable or wire bundle). The resistance
of the coil should be as small as possible without shorting out the
PRBS-code generator (i.e. produce maximum sustainable current).
[0055] The output coil also should fit snugly around the cable, and
the number of windings should be large. As the number of windings
increase, the voltage induced in the coil increases, and the
induced current decreases. The optimal number of windings will be
determined by the current requirements of the circuitry reading the
output coil voltage. Preferably, the input and output coils will be
integrated into one unit (but electro-magnetically isolated from
one another) that can be simply hooked on a conducting
structure.
[0056] Some applications where the method and system of the present
invention are particularly effective include testing of wiring and
cabling, particularly aircraft wiring, for chafing or insulation
defects, detecting defects, such as cracks, corrosion and abrasion,
and/or leaks in piping, detecting leaks in piping and detecting
anomalies in various elongate conducting structures, such as
fencing breaches.
[0057] Inasmuch as the present invention is subject to many
variations, modifications and changes in detail, it is intended
that all subject matter discussed above or shown in the
accompanying drawings be interpreted as illustrative only and not
be taken in a limiting sense.
* * * * *