Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
277~-233D
This is a divisional application oE copending Canadian patent
application Serial No. 379,292 filed on June 9, ]981.
This invention relates to new and useEul electro-optical methods and
apparatus using photodetector arrays for measurement of parts and other ob-
jects in ;ndustrial environments. Several particularly useful optical embodi-
ments are illustrated together with desirable electronic circuitry for improv-
ing the accuracy oE practical systems of this type.
All embodiments disclosed utilize detection oE one or more points in
light patterns to effect the measurement. Such patterns can be formed using
ultra-violet, visible and in:Era red light, and typically include profile edge
images of parts, re:Elective images of part surfaces, diffraction patterns pro-
duced by one or more edges of a part and images of spots projected onto a part.
The electronic technique disclosed is primarily aimed at improving
resolution beyond the limited number of array elements available, negating
effects of inter element sensitivity differences and compensating for light
pc>wer variations both in time and spa-tially across the array. All of these
Eactors are important in obtaining accurate performance of such systems in the
industrial and other non-laboratory environments. "Real time" operation is pro-
vided, allowing high part rates or large m1mbers of measuremellts to be taken.
2n According to a first broad aspect of the present invention, -there is
provicled a method of sorting cylindrical parts, said method comprising provid-
ing movelnent oE the cylindrical parts along a predetermined path of travel pastn station at ~hich ~ photodetector array is located;
cletecting that the part is in a precletermined position relative to
tlle photo-letecto-r array using a part present sensor;
arrangillg the photocletector array at an inclined angle with respect
to the Irclth oE-travel oE the parts and thus with respect to the parts traveling
nlong thtlt path in or-ler to permit viewing oE an end of each part;
ll.S. No. 1()3,290 DIV I -1-
.' ~
pulsing a light source, responsive to saicl part present sensor, so as
to :Eorm an image of said part on said photodetector array;
determining at least one dimension of the part from points in said
image;
comparillg the dimension so determined with a stored value of said
dimension; and
sor-ting the parts into categories based on said comparison.
According to a second broad aspect of the present invention, there is
provicled a method of sorting cylindrical parts, said method including the steps
iO oE:
providing movement of the cylindrical parts past a station comprising
a photodetector array using a vee-shaped belt in which the parts are received
and at least a portion of which is light transparent;
detecting that the part is in a position to be measured using a part
present sensor;
pulsing a light source to direct a light heam therefrom through the
light transparent portion of said track to form an image of the part on said
photodetector array;
determining at least one dimension of said part from points in said
image;
comparing the dimension so determined with a stored value relative -to
saicl dimension; and
sorting the parts into categories based on said comparison.
Accorcling to a third broad aspect of the present invention, there is
providecl a me-thod of sorting parts including the steps of dynamically passing
eacll par-t past a station comprising a photodetector array, de-tecting that the
part is :in position to be measured using a par-t present sensor, pulsing a light
-la-
source to freeze the image o:E said part formed by a lens onto said photo-
detector array; determinillg -Erom the output o:E sai.d photocletector array one or
more dimensions of said part from points in said image; comparing the dimen-
sions so obtained to stored values and actuating a gate, as required, to sort
parts into categories, said determining step including scanning said array to
produce a stepwi.se variable signal comprising a series of pulses, filtering
said stepwise variable signal, differentiating the -filtered signal to obtai.n
-the second clerivative thereo:E for said points in saicd image and detecting the
zero crossing of the differentiated signal.
Accorcling to a fourth broad aspect o:E the present invention, there is
provided apparatus for sorting cylindrical parts, said apparatus comprising: a
photodetector array located at a station along a path of -travel, means for pass-
ing the parts past the station at which said photodetector array is located,
means for detecting that a part is in a predetermined position with respect to
the photodetector array, the photodetector array being arranged at an inclined
angle with respect to the path of travel and thus with respect -to the parts
traveling along said path o:f travel in order to permit viewing of an end o-f
each part, a light source, means :for pulsing said light source to form an image
of said part on said pllotodetector array, means for determi.ning at least one
?O dimension of said part from points in said image, means for comparing the dimen-
sion so cletermined with a stored value of said dimension, and means Eor actuat-
lng a gate, as recluircd, to sort the parts into categories.
Acco:rd:ing to a fiftll broad aspec-t of the present invention, there is
rov:kled al~lnrahls Eor sorting parts i.ncluding subs-tantially cylindriccll por-
tiOIls, sa:icl al)l~aratus compris:illg: a photodetector array located at a sta-tion
alollg a pa-th o-f travel; means ~for provicling movement o-E a part past the sta-tion
at wllicll said pllotodetector array is located, said movement providing means in-
-lb-
~2~ 0~
cluding a vee-shaped track in which the part is receivecl, and at least a por-
tion of which is light transparent; part position detecting means for detecting
that the part is in position to be me2sured; a ligh-t source Eor, when actuated
clirecting light through the transparent portion of said vee-shaped track; a
lens Eor -focusing light from the light source onto said photodetector arrayi
means Eor pulsing said light source to form an image of the part said photo-
cletector array, means responsive to the output of said photodetector array for
cletermining at least dimension of said par-t from points in said image, compar-
ing meLIns Eor comparing -tl-e dimension so obtained with a stored value o-E that
1() climension, and means responsive to said comparing means Eor actuating a gate,
as required~ to sort the parts into categories.
According to a sixth broad aspect of the present invention, there is
provided apparatus for sorting parts, comprising a photodetector array, means
for passing a part past said photodetector array, means Eor detecting that the
part is in position to be measured, a light source, a lens for forming an image
of said object onto said photodetector array, means for pulsing said light
source to freeze the image of said part formed by the lens onto said photo-
detector array, means for determining one or more dimensions of said part from
points in said image, means for comparing the dimensions so obtained to stored
valucs, and means for actua-ting a gate, as recluired, to sort parts into
categories, said determining means including means for scanning -the outputs oE
tl~e -inclivkl-lal elements oE the photodetector array to procluce a stepwise vari-
nl)Lo sigll.ll comprisiilg a series o-E pulses, low pass Eilter mealls Eor Eiltering
snid sigllaL, cLiEEerentlcltin~ eans for diEEerentiating said f;ltered signal to
obtain the second clerivative thereo-E for said points in said image and means
~or cletoctillg the ~ero crossing of the diEferentiated signal.
Accordillg -to a sevent}l broad aspect oE the present invention, there is
-- lc--
3~5
provided apparatus accorcling to Claim 31 wherein said comparison means is a
micro-computer .
The invention will now be described in greater cletail with reference
to the accompanying drawings, in which:
Figure l illustrates a typical arrangement according to the invention
for carrying out measurements by electro-optical techniques in an industrial
environment;
-ld-
Figure 2 is a block diagram illustrating a signal processing
circuit used in the invention;
Figures3a to 3f are timing diagrams showing wave forms generated
at different parts of the circuit of Figure 2;
Figure 4, appearing on the same drawing sheet as Figure 1,
illustrates the measurement of a wire using a hand held "gun", according
to the invention;
Figure 5 is a block diagram similar to Figure 2 but showing a
s:ignal processing circuit used in the Figure 4 embodiment;
1() Figure 6a to 6d illustrate various stagcs oE the processing
of electronic signals according to the invention for a diffraction Eringe
application;
Figures7a to 7d illustrate various stages of the processing o-f
electronic signals according to the invention for a light to dark transition
of an optical edge;
Figures 8a to 8d illustrate various stages of the processing of
electronic signals according to the invention for the same optical edge that
has been defocussed;
Figure 9 illustrates the use of the invention to see defects on
automotive camshaft lobes;
Figures lOa and lOb are schematic elevational and plan views,
respcctively, illustrating another embodiment of the invention used for
sort:illg :Easteners and the like;
Flgure l:L is a block dlagram illustrating application of the
;Invcnt:ioll to 2 dlmensional lnspection;
-- 2 --
ligure 12 is a partial schematic v:iew illustrat;ng an embodiment of
the invention used for determining the position of automotive sheet panel edges;
Figure 13a illustrates schematically the application of the i.nvention
to determi.ning dents on sheet metal panels; and
Figures 13b and 13c illustrate schematically images obtained :Eor a
good panel and a bad panel, respectively.
Figure 1 illustrates a typical measurement according to the invention.
An object, in this case a cutter insert, 1 is illuminated by light -Eield, 2, o:E
lig}lt source 3. The image ~t (inverted) of the cutter edge is forrned by lens 5
I() onto matrix photo diode array 6. Each vertical row of said matrix is analyzed
by circuit ~ to obtain either a digital display 10 of cutter dimension in any
section, an accept/reject comparison with light output 11 or a visual display
12 of the cutter edge or outline thereof on a TV screen. Generally windows l~t
and 15 in housings 16 and 17 are also present to protect the optical elements.
Electronic processing circuitry, described below, is very important
as the accuracy of a system such as figure 1 is totally dependent on the
abi.li-ty of the analyzing circuitry to ascertain where in space the image is on
the diode array, in the presence of diode array sensitivity differences, clirt
on -the windows, l~t and 15, and non uniformity of the light field 2 across the
2n object edge.
The electronic processing now described applies equally to linear
plloto detector arrays, and circular arrays as well as matri~ arrays such as 6,
~2~)~9~5
on a one axis line hy line basis. Othér techniques for two axis image edge
detection and enhancement will also be disc].osed.
At the present state of the art in building photodetector arrays
physical limitations are encountered in producing small array elements which
in turn therefore limited the use of arrays to coarser measurements and
inhibited working in field applications. The array size limitation imposes a
limit on the spatial resolution achievable in practice, which falls far short
of the theoretical limits of the optical imaging elements.
Furthermore, in these devices element to element sensitivity
variations are unavoidable due to the very complex nature of the array manu-
facturing process. This factor coupled with variations in the light field lead
to gross errors in photodetector array based optical measurement systems.
The disclosed invention overcomes all of these limitations and makes
high accuracy measurements possible in areas heretofore closed to automated
optical gaging systems.
By their nature, discrete element photodetector arrays will only
provide discrete spatial samples of the light field. The theoretically
obtainable spatial resolution in detector array systems is limited to the
physical dimensions of one individual array element, when simple thresholding
techniques are used on the array output signal. In practice, these
theoretical limits can only be approached, but never reached.
To overcome the limitation of the finite element size of detector
arrays one has to consider the following. When an array is sampled
sequentially at a precise -frequency the video signal produced will be a time
domain representation of the optical signal in space. Once the signal is in
the time domain one can process the signal by analog electronic frequency
9os
domain circuits. Low-pass filters and differentiators provide the necessary
frequency domain signal processing capability.
A block diagram embodiment is shown in figure(2). The video signal
coming from photodetector array ~a) is low-pass filtered by a linear phase
filter (b) such as a ~essel filter, then differentiated twice by ~c) and
(d) following the signal is zero crossing detected (e). The time interval
between the start of the array and a zero crossing of the differentiated
signal, or the time interval between the first and subsequent zero crossing~
is measured by a precision timer (f). The time base signal for the timer and
1~ the clock pulses for the sampling of the diode array is produced by a highly
stable clock generator (g). Typically the location of the image edge is
then fed to a microcomputer as shown for comparison with stored values of
master parts or the like.
The importance of using this precision timer to obtain spatial
resolution enhancement is shown in Figure 3.
Figure 3a indicates a magnified view of a transition from light to
dark as would be observed at the edge of a part. The horizontal axis
indicates time (which is proportional to position on a photodetector array)
and the vertical axis indicates a voltage which is proportional to the
intensity of the light falling on the array. The edge of the object is
defined by the point of inflection of the signal.
The precise location of the image edge cannot be readily seen
because of the finite size of the array elements.
Passing this signal through a low pass filter provides an interpolated
signal that reduced the effect of the sampling frequency (figure 3b).
Figures 3c and 3d show the signal after being differentiated twice.
9~
The point of inflection that represents the edge of the image is now
indicated by the position that the second derivative crosses zero. The zero
crossing detector shown as E in figure 2 will de~ect the position of the image
edge.
It is in the process of communicating this edge position to the
computer that the timer precision of F (in Figure 2) becomes important.
Since the goal is to determine the position of this image edge, a counter or
clock timer must be used to keep track of which element of the photodetector
array the zero crossing refers to.
When the zero crossing detector indicates that the edge of the
image has been found, the value of this clock is stored by the timer (F). If
this clock is no~ pulsing very often, the image edge will be found but the
position won't be stored until later (which means further away).
Consider Figure 3e. The period of this clock is the same as the
element to element period of the array. In this case the image edge will be
found to be on clock pulse 3 or 4 depending on the exact position of the edge.
This corresponds to a spatial resolution of only l part in 7. (Detector arrays
with many more elements are available but this 7 element array is sufficient
to illustrate the technique)~
2.0 If a higher frequency clock is used (Figure 3f)~ the occurrence of a
zero crossing is checked far more often. Therefore there is less delay
between the time a zero crossing occurs and the ~ime its existence is detected.
In this example the clock timer is 5 times the frequency of the element rate
of the array. Now the image edge will be found to occur at clock pulse 16 or
17. This corresponds to a spatial resolution of 1 part in 35 using the data
from the same photodetector array. Another way of looking at it is that each
array element has been broken into 5 parts, giving results similar to those
of an array with 5 times as many elements. It is important to note that the
high speed clock does not have to be an even multiple of the array clock
~although it is sometimes convenient). It is only necessary that the high
speed clock be synchronized with the beginning of the video scan to provide
repeatable results.
This technique has been shown to locate the edges of an image 64
times more accurately than would be possible using a low speed clock.
The use of a well designed low pass filter is crucial to the
operation of the embodiments shown. The filter has two specific purposes
in the electronic lineup. One is to remove the sampling frequency components
from the video signal, the other is to provide an instantaneous weighted
average of several array elements. By producing this instantaneous weighted
average, the array element to element sensitivity varia~ions (typically
5-10%) can be overcome, thereby improving the consistency of the readings.
This also provides a necessary interpolation between elements.
The elimination of the sampling clock noise from the signal is
required because the differentiator following the filter would respond to
the clock frequency components present in the video signal.
Depending on the optical signal being processed and the particular
aspect of the signal being sought, one uses single or double differentiation
before zero detection to arrive at the signals to be timed by the precision
timer.
Double differentiation is used when the inflection points are south
in optical signals, such as the light/dark edge transition of an image, or
multiple edge transitions found in images of bar codes, or images of surface
~Z~)~9-~S
flaws. Also diffraction fringes and triangulation spots can also be
analyzed by determining inflection points as well.
The very high resolution timer ~igure 2(g~ utilized in the system
allows the detection of the precise location of the features sought in the
optical signals to great accuracies. Tests have shown, that detec~or array
element spacings can effectively be subdivided into 64 parts with 1 count
error. This very high resolution allows the construction of optical instruments
with accuracies formerly unattainable.
The system as described here also provides optical power compensation
by its nature. Variations in light intensity have no effect on the locations
of image edge points detected. Typically the dynamic range is 8 to 1 in
intensity when the array element resolution is enhanced by a factor of 8.
This tolerance to light level variations makes the measuring instrument quite
insensitive to light field variations due to changes in reflectivity, aging of
light sources, settling of dust on the imaging lenses etc.
The double differentiation technique operates on finding the exact
location of the inflection point in the light/dark transition of imaged
edges of solids. This inflection point does not change its location as the
light field is varied and double differentiation coupled with zero crossing
provides a system which is quite insensitive to optical power charges. The
added benefit of this system is the insensitivity to defocussing. A
considerable amount of image degradation has to take place before the
capability to detect an edge of image is lost.
Note that the concepts above apply equally to reflective (front
surface) illumination, and are useful for detection of surface defects as in
our Uni~ed States Patent No.4,326,808 which issued on April 27, 1982.
~z~
The analysis concepts described here are useful also in determination
of diffraction pattern fringe locations, either using the first or second
derivative of the array signal.
A variation on the disclosed electronics is also useful in real ~ime
processing of diffraction patterns used in industrial measurement systems an
example of which is described as follows. Consider Figure 4 which illustrates
measurement of wire diameter with a hand held "gun", 50, according to the
invention.
As shown, diode laser 51 illuminates via lens 52 and mirror 53 the
LO edges 55 and 56 of fine wire 57, typically coated magnet wire ln~l or less in
diameter. The resulting diffraction pattern 58, resulting -from interference
of diffraction waves emanating from said edges is detected and scanned by
photodetector array 60. In this case a type Reticon 512C photodiode array.
Mirror 61 is used to deflect the light in order to make a compact housing.
This circuit on board 62, whose block diagram is shown in Figure 5,
has been designed to process diffraction fringes falling on the diode array.
The output of the circuit is a set of two binary numbers, one is the number of
fringes detected the other is the distance measured between the first and the
last detectable fringe. The average fringe frequency is arrived at when the
numbe:r of Eringe spaces is divided by the distance as measured by ~he diode
a.rray. From this and two other constants the wire diameter can be calculated.
The basic diffraction equation is:
W = nR~ where W = slit width
Xn n = fringe number
R = distance between slit and
detector
W = R~B or b - n ~ = wavelength of light
Xn (HeNe laser ~=
6.328nm)
B = fringe frequency
where
B = no. of fringes detected
distance value
The intensity profile is I - A sin 2a
a2
~liS equation gives a cyclic intensity variation, where
I = 0 when a = ~, 2 ~, n ~. In other words~ the minima ~or
fringes) are located at regular intervals. The spacing can be measured
by simply locating the position of the first and last fringes on the diode
IU array, taking the distance between them and counting the fringe spaces in
that distance.
There is another implication in the above equation namely
by a simple differentiation of the video signal and zero crossing detection
the minima can be located accurately and quickly.
The spatial resolution of the diode array is enhanced by the use
o~ an averaging low pass video filter and a precision timing clock tha-c
operates 4, 8 or 16 times faster than the diode array clock. When a minimum
point is found its position can be determined 4, 8 or 16 times more precisely
than by the conventional clocking method.
The description of a dif~raction system is beyond the scope of
this discussion, however, some of the advantages of diffracti~n are given }lerc:
(a) There is no theoretical limit of resolution in a diffracting
systcm (l~in. resolution llas been achieved).
(b) No lens needs to be used, so a variable element is el:iminated
from the opticnl train.
(c) Magnification varies witll the R distance, because of this a
- 10 -
~L2~
diffraction system is less sensitive to axial position variations than an
imaging system.
This circuit described in this invention has been designed to
operate with a boxcar type video signal generated by a *Reticon RC 100B board
or equlvalent.
The standard pin edge connector of this board sits on the Diffracto
CPU bus and is memory mapped. On the opposite end of the board there is a
connector to provide junction to a round cable coming from the RC lOOB board
in the optical head. This connector carries the video blanking, array clock
and array start signals along with a flash trigger signal and ~ ground pins.
The intention is to have only signals and ground lines in the cable and to
have the supply voltages independently connected in the optical head.
The boxcar video signal is filtered to remove the clock frequency
and any odd/even pattern noise in the signal. The filter utilized is a 6 pole
Bessel low pass filter with its cut off frequency set at 1/10 of the array
clock frequency. Operating this way the filter keeps the clock frequency more
than 100 dB below the signal level. The advantage of using a Bessel filter
is that it does not "ring" or produce unwanted harmonics. It also acts as an
averager for several diode array elements at any one time, following a
weigllted window :Eunction. This feature, coupled with a sys-tem clock operating
~() at a Erequency several times the diode array clock frequency will allow the
rosolut:ion to exceed that of the element resolution level inherent in any
cllodo array.
The diffraction pattern minima are found by differentiating and
cl~ecting the ~ero crossings in the filtered video signal.
k Tradc Mark
~2g~ S
At the first minimum in the diffraction pattern a fast counter is
started. The ratio of the counter clock frequency to the array clock
frequency is referred to as the electronic magnification factor (EMF).
After the first fringe minimum, succeeding millima are used to latch
the instantaneous count into two output ports. After the last fringe
minimum encountered on the array, the number remaining in the latches is the
counts between the first and last fringe.
A separate counter is used to count the number of fringe spaces
encountered on the array scan. The total number of counts and the number of
fringe spacing can then be accessed by the resident computer system.
~hen the diode array is in saturation, the digital latching hardware
is disabled to prevent latching the count on false fringes. Due to the
fringe envelope, the outer fringes on the array may be too small to give a
reliable reading. A small fringe detector circuit disables the latching
hardware when the fringe maxima fall below a set threshold.
A monostable is used to delay gaging on the scan start allowing
for filter settling time.
As soon as a DATA REQUEST occurs, a start pulse is sent to the
array for the DUMP cycle. At the end of the dump cycle a flash monostable
is triggered. This monostable output is available on the connector for flash
triggering (pulsed lasers). After the flash window, another start pulse is
sent out to the array for the READ cycle. Gaging takes place during this
array scan.
The data from circuit 62 is sent to a microcomputer 65 (typically
nn *Intel 8085) and processed to display wire diameter in inches, microns, etc.
alld disp]ay 68. Accuracy is in the area of 1-5 millionths of an inch
* Trade Mark
(depending on the number of array elements utilized) and is superior to
all other known wire gages.
The system described is also ex*remely linear, with less than 10
millionths error over a whole range of diameters .0005 to.O20" . In
addition the same electronic approach can be used on diffraction patterns
resulting from two edges or a cylindrical part and an edge, as typified by
United States patent 3,994J5~4 by one of the inventors here.
When diode laser 51 is operated continuously or a gas laser is used,
reacllngs at rates up to lOOO/second can be obtained. This is generally fast
enough to freeze the wire position with no pattern blurr. However, this
can absolutely be accomplished by using lens 52 to focus undiffracted
radiation at the plane of the array, thereby creating a quasi static pattern,
and by utilizing a pulsed diode laser 51 to freeze the pattern with a short
stroboscopic pulse. This pulse can also be triggered by a separate detector
65, and circuit 66, used to indicate that the wire is in position to be
measured - a particularly desirable feature in a hand-held unit.
Naturally the unit does not have to be hand held and can be fixed
in place and included in a control system to feed back data to control a wire
coating machine for example.
Analog detector 70 positioned on one of the ramps of a Eringe in
the diffraction pattern and comparison circuit 75 incorporating a iligh pass
EiLter can also be used to detect "bumps" in magnet wire coatings. Such
bumps represent typically 20% or greater diameter cilange and cause a violent
but nlomentary sllift in the diffraction fringe location which is immediately
detected by circuit 75. One bump can be detected or a count of bumps per meter
cnn be obtalned v:ia the microcomputer.
~3~
The stages of the processin~ of the electronic signals
achieved by this invention is demonstrated in figure 6,
7 and 8 for a number of differen-t ap?lications. rigure
6 illustrates the process for a diffraction fringe app-
lication, figure 7 for a light to dark transition of an
optical edge and figure 8 Eor the same optical edge that
has been defocussed.
Fig. 6 top trace shows the video output signal of a photo
detector array, the optical signal generating it is a
single slit diffraction pattern. The spatially sampled
nature of the signal is clearly visible in the dot pat-
tern (trace a) that makes up this signal. The centers
of the fringe maxima are saturated. Without further
processing any resolution of space would be limited to
1 element represented by each of the dots.
Trace (b) is a low-pass filtered version of the first
signal. The continuous nature of the optical signal
at this point is clearly restored-and there are definitely
'no discontinuities in the signal. The apparent shift
to the right of the signal b, c and d i5 caused by the
delay in the low pass filter and the dif~erentiators.
The Bessel filter is chosen because it exllibits a con-
stant time delay which is not dependent on the video
signal. Since this time delay is constant it can be
easily recounted for in the system calibration. The
continuous nature o~ the optical signal at this point
is clearly restored and there are definitely no discon-
tinuities in the signal.
Trace (c) is the filtered video signal after differen-
tiation. l'he negative to positive goiny zero crossings
in the di~fcrentiated signal coincic'e with the filtcred
..
s
signal-minima. Tr~ce (d) shows the output of the zero
crossing detector, the negative going edges accurately
locate the relative positions of the diffraction pattern
minima.
Fig. 7 shows array signal output traces, indicative of
light/dark transition of a greatly magnified edge of an
object projected onto an array using a lens. Again, as
clearly seen in trace (a~, spatial sampling of the array
destroys spatial continuity, which is restored by ~he
constant time delay low-pass filter as seen in-trace
(b). The filter also provides interpolation between
sampled points in space. Trace (c) is the first diff-
erential and trace (d) is the second differential of
trace (b). The inflection point in the light to dark
transition is fixed in space and can be found very pre-
cisely using the double differentiation method. Zero
crossing occurs in the second differential at the in-
flection point and this point does not shift with
changes in light intensity, therefore, it provides an
automatic means of optical power compensation.
Fig. 7 shows a focussed image trace whereas Fig. 8 shows
equivalent traces for an image somewhat out of focus.
As seen from the signals zero crossing detection will ~;
importantly still give the location of the edge in que-
stion although magnification changes can be experienced.
In the embodiment of figure 1 the disclosed circuit plus
the solid state digital array are crucial to providing
a workable high resolution device capable of fulfilling
modern industrial requirements. For example it is com-
monly required to set cutters to accuracies o .0001"
or better which can readily be inspected in the apparatus
- 15 -
~.
g~
of figure 1.
A typical detector ~or the cutter application could be a
General Electric (GE) TN2200 128 x 128 element diode
array with elements on 1.~- x 1.2 mil centers, and a lens
magnification o~ 2O1, in this case the image of the cri-
tical edge of the cutter tip will lie on the array however,
with a'-resolution of .0006" per element which is insufE-
icient to serve most applications. Wi~h the application
of the disclosed invention to find the edge image, the
image resolution easily becomes 20 millionths,of an inch'
No drift or non linearity occurs due to the digi,tal diode
array position.
In the figure 1 example the cutter insert 1 is locked
in boring bar 20, ~esting in Vee block setup fixture 21.
It is now of interest to consider tying in of optical
x, y, and 2 axes provided by slides 25, 26, and 27 each
witn its own position encoder ~not shown for clarity~.
Provision of such axes is similar to many optical coor-
dinate projectors whicn may well be displaced by device
here disclosed.
From the readings obtained, one can project the cutter
contour and location greatly magnified in the scan axis
onto a video displayO This is generally an analog CRT
which has its own error unless it too is a digital array
device such as a LCD flat display not yet widely avail-
able.
More appropos but more expensive is to automatically
compare the cutter edge data to limits. ~or example,
th~ cutter ~oint can be ;mm~d; ately found by comparing
successive scans, as can cutter angle. Since resolution
- 16 -
is very h~gh, cutter defects such as chipped out areas can also be detected as
well.
The considerable resolution enhancement provid d allows measurement to
high resolution (eg. .0001") over a large field of view. For example, over
1" field a co~monly available GE TN 2500 250 x ~50 element array with the dis-
closed electronic technique easily provides sufficien~ resolution in the scan
axis and via computer augmented processing in the other axis as well (see
Figure 11).
Lens errors over the field can be corrected in the microcomputer if
desired which avoids the cost of very expensive flat field distortion free
lenses. This can be done by storing points in a precision xy gratin~ placed at
the object position and interpolating test data obtained between these stored
known points. This also ~akes ou~ residual light filed errors although the
disclosed circuitry can account fo-r almost all such errors.
The above resolution of .0001" over a 1" object fieid is far better
than obtainable with even the largest optical projectors and accuracy is "loc-
ked-in" due to the driftless and digital solid state array. Range however
can be vastly increased in an automatic manner by tying the readings obtained
in to the slide readouts shown.
Figure 9 illustrates the use of the circuitry to see pits, scores and
other defects on automotive camshaft lobes eg. 99. The basic defect analysis
processing differ from that disclosed in United States Patent 4,326,808,
in that here the defective edges are detected by the double differential cir-
cuit and their width determined between inflection points or by the number of
~uccessive
~2~ CIS
scan lines on which a flaw is detected. ~-lowever, -the
optical apparatus here described may also be used with
other circuitry such as that of the referenced a~plication.
A linear diode array 100 such as a ~eticon 256C on which
the image 101 of the cam lobe a~ial strip 102 viewed by
the array is formed by lens 103 typically of large number
for good depth of field~ In practice this is accomplished
using a lmm diameter objective lens to limit the angular
aperture. Light source 110 comprised of continuous arc
light or numerous illumination individual sources arranged
in an arc illuminates the cam lobe from multiple angles
such that sufficient light can reach the array during
all positions of cam angular rotation.
With such arrangement relatively little variation in light
power returned form the surface exists and this can be
accounted for by the ~rocessing circuitry as described.
Typically 100 scans/sec. or yreater of the diode array
may be achieved.
~ote that this flaw detection technique can also be used
on other curving and focal depth varying objects. Of
particular interest are ball bearings, raceways, and cav-
ities such as molds etc.
Figures lOa and lOb illustrate another embodiment of the
invention used for sorting fasteners and other parts of
all sorts. Tne particular embodiment is set up for cy-
lindrical type parts such as bolts, dowels, rivets, etc.
A bolt 200 ~ed by feeder 201 slides down Vee track 205
havin~ translucent section 206. A matrix photodiode
arr.~y 210 is positioned above the track, with a ~enon
flash lamp ~12 below. As the bolt slides down ~he track
- 18 -
9~5
(generally oricnted thrcadcd end first t~laJIks to fee~der
-tooling) the bolt end tri2s a light emitting diode and
detector part present sensor 220 which fires the flash.
Tne light from the flash travels through the teElon track
tn2reby freezing the bolt ima~e or a portion thereof on
the array by virtue of objective lens 221~
The scan axis of t'ne array lines is typically oriented
across the bolt diameter ie. perpendicular to the long-
itudinal axis of the bolt. Thus the real time resolution
improvement resulting from the electronic techniques of
the disclosed invention increases the diametral measuring
accuracy.
A microcomputer controller 240 is utilized to compare
multiple diametral sections to stored acceptable values
for the bolt plus compute the overall len~th, threaded
length, head height, washer thickness (if any) and captive
washer presence (if any) from se~uential line scan data.
Optionally the number of threads can also be counted as
well.
After the comparison, the part is accepted or rejected
by blow-off air solenoid 250 as shown.
A detector 260 can be located in the image plane adjacent
to the matrix array to turn off the flash when a normal
energy level has been reached, thereby compensating for
dirty windows or track.
The Xenon flash source can optionally be replaced by a
bank of pulsed solid state sources such as diode lasers
or LEDs, 270, as shown. The circuitry here described,
- 19 -
unlike fixed threshold types can be used with uneven light
flelds as result from such use of discrete sources.
~lachines built in t'nis manner will make significant- pro-
ductivity improvements by providing prequalified parts
to assembly machines. Indeed the sensin~ units-here dis-
closed can be mounted right in the feed tracks of said
machines~
.,
Diametral accuracies of .002" uslng an inexpensive ~ ~
22~0 array system are possible on 2" bolts. Note that
a special part present detector 275 (shown in Fi~ure lOb)
is required upstream to dump charge from the array before
the flash. This extra detector is necessary to provide
repeatable triggering ~or rapidly moving parts.
The photo detector array must be continuously scanned
to keep the array cleared of charge due to stray light
and leakage current (on the detector array itself). The
~,scan rate of the Tl~ 2200 camera is 20 scans per second.
Since the parts are travelling at a high speed, the array
can miss the part completely if it waits until the end
of the next scan to fire the flash. Firing the flash`
during the scan is undesirable because the format of the
video signal makes the processing o it dificult. ~:~L
Using a second part present detector (275) solves the
problem because it gives the array advance notice that
a part is approaching. Given this infor~ation, the array
can finish its current scan and stop scanning at the end
of scan. When the part reaches detector 220, the flash
can be fired and the data read instantly. The time that
the scanning is stopped is not sufficient to allow sign-
i~icant build up of stray light or leakage current.
- 20 -
S
A special display with freeze memory is also provided
witb cursors to set up diametral and longitudinal com-
parlsons.
For inspeckion of bolt head dimensions and washer inte-
grity plus large head cracks and the like, a second solid
state matrix array camera 280 can be used, positioned at
a slant angle to the track to view the head area end on.
For general washer and head inspection the back lighting
provided by the same flash coming through the track su-
fficies. Optionally additional light sources can be
provided suc~ as 285 reflecting o~f the bolt head~
A base 290 is usually provided for the above com`~onents,
as are protective housings such as 291 and 292 with win-
dows 293 and 294.
Note that on long parts the lens magnification can be
set up such that only the head end is viewed since the
flash occurs at a precise point of triggering if a pre-
cision part position sensor such as shown at 295 in Figure
10b comprised by lens and detector with slit aper-ture
is used. This freezes the head end allowing one to read
overall length, threaded length and ~enerally all other
meaningful variables as well.
The Vee track is inexpensive, constrains well the cylin-
drical parts, and is particularly useful for bolts where
the head is o~ significantly different diameter. Simi-
larly a translucent Vee section belt with motor drive can
also be used, which allows higher part rate due to speed
consistency.
Finally a motorized Elat bel~ of mylar or other trans-
-parent or translucent material can also be used to tran-
sport the part witn the pa~ constrained by guide rails.
Indeed an opaque belt can be used if the light source
and camera i5 horizontal-.
~ext disclosed lsee ~igure 11) is an extremely useful
aclditional application of the circuit technology dis-
closed to 2 dimensional inspection using a matrix array,
a sequen-tial line~scan of a moving image (either due to
motion of tne part or a sweep of its image past the linear
array in question), or for processing of sequential scans
of a circular array like a Reticon 720 but having mult-
iple rincJs or multiple elements (such arrays are not yet
commercially available).
In brief, the technique for two axis mensuration and a
matrix array is to read the lines of elements, line by
line and determine the edge locations using, for example,
the circuit here described. This defines the edges in the
direction of the scan lines. The second dimension is
obtainecl by reading the information column by column,
which is at right angles to the lines of the array, and
process the information by -the disclosed electronics.
l~he technique here is to additionally read each element
of each scan into a gray level digital memory (eg. 2.56
levels, or 8 bits) and then read out each equivalent
array column o the memory into the same circuit dis-
closed (eg. that of fig. 2 or 5). In this manner the
point o edge location in 2 planes is dete~nined, which is
o value in generalized measurement of 2 dimensional images
of objects. This technique a]so works on 2 axis diff-
xaction patterns such as those produced by square apertures.
- 22 -
Video data from the pho-to detector array (~) is ~assed
through the ~ideo switching network (~) to send data to
spatial resolution enchancement system (L). The spatial
resolution enhancement system locates the image ed~es for
each array line and the location of these edges are
stored in data storage element (M).
At the same time, the video data is converted to digital
form by the analog to digital convertor (~ The video
signal from each array element is converted to a number
that is proportional to the light falling on it. This
series of numbers (one for each array element) is stored
in the digital buffer storage (I). The data is stored in
the order of the array lines (ie. line 1 element 1, line
1 element 2..... line 2 element 1, line 2 element 2, etc.).
After the edges for the lines have been stored (as men-
tioned previously) the video switching network (K) is set
to receive the signal from the digital to analog convertor
(J). The image data is recalled from the digital buffer
storage ~I) one element at a time and converted back to
analog form. However, when the data is read back, it is
read in the order of the array columns (ie line 1 element
1, line 2 element l..... line 1 element 2, line 2 element
2 etc.)
,.
Now the spatial resolution enhancement system (L) locates
the image edges for each array column and the location
of these edges are stored in the data storage element
(M). ~y using this technique, the same resolution ench-
ancement system can be used for both the vertical (col-
umns) and hori~ontal (lines) directions of the image array.
- 23 -
~2~
By ch~nging the order that the elements of the image data are read
- from the digital buffer storage ~I), the location of the edges on diagonal lines
may also be ~ccurately found and s~ored. Again the processing elements L and
are used to locate and store edges for all directions that may be scanned by
storage unit I.
Finally it is noted that the above disclosed invention, par~ -u]arly
in its 2 axis form is of considerable value in determining the position of ob-
jects in space. Particularly considering the use of either shadow or reflected
objec~ images as in Figure 10 or 9 respectively, one can immediately derive ben-eficial results in "vision" systems used for robot control. ~te accurate delin-
ia~ion via the invention under all ~onditions of holes, edge outlines and other
features of parts is essential for proper operation in the industrial environ-
ment.
Figure 12 illustrates an embodiment of the invention used for determin-
ing the posi~ion of automotive sheet metal panel edges in space, when attached
to a robot. The sensor utilized operates on an optical triangulation principle
which uses 2 or-thogonal axes of measurement znd inCOTpOrates the desirable cir-cuit aspects such as described heretofore.
As shown, the position in x and y of sheet metal panel 30 is simultan-
eously determined by projec*ing spots 303 and 304 onto its surface by beam pro-
jectors 305 and 306, typically comprising diode laser 307 and focusing lens 308.
- 2~ -
Said spots are imaged by lens 3I0 to form spot irnages 315
and 316 on linear photo detector array 320.
In operation, tne 2 axis sensor ensemble 325 is moved
near the door edge by robo-t arm 330 (typically an ASEA
or Cinncinatti Robot) and sheet panel position obtained
through reference to the robot co-ordinate axes.
.
This is an independent dual channel system and can be
operated in one axis with either 305 and 306 extinguished.
To separate the readings in x and y therefore one need
merely pulse the source 306 and determine the position
of the y axis spot on the diode array, then shortly
thereafter pulse the x spot from 305 and do likewise.
The spot image on the array is proportional to location.
To determine the spot image location on the array, a cir-
cuit such as that fig. 5 is utilized to find the first
derivative of peak location of the spot. Alternatively,
a circuit like fig. 2 can be used to find the inflection
points on one or both sides of the spot and the centroid
computered the microco~puter or hardware circuitry.
This circuit approach particularly using high speed
timer with a clock rate in excess of the array rate is very
advantageous for increasing the resolution of sucn tri-
angulation systems which typically are now very sensitive
as only low power magnification of the lens system can
be used.
The above 2 axis proximity sensing system, besides being
use~ul for measuring sheet metal contours and edge loc-
~tions, can also be used to advantageously provide feed-
- 25 -
~2~
. .
back signals to robot control systems as a function of
-part location in 2 orthogonal range axes. Combination
of this with the previous embodiment providing 2 axis ~
mensuration of images in the plane perpendicullar to the
lens axis, provides a complete 3-4 axis measurement and
control capability.
All of the foregoing embodiments are representative of the
scope of the invention and many other variations are
possible.
~s mentioned above the invention can also be utilized
for measurement of edge locations of bars in bar codes,
using in general reflected light off the code printed or
etched on the surface of a part or a tag put thereon.
In the bar code case, the invention gives a very desir-
able freedom from returned light power variation, all
across-the field of view of the diode array/ an important
,point since on long bar codes, such as code 3S, it is
very difficult to have uniform lighting on all bars us-
ing inexpensive sources (such as tail light bulbs) and
is further made difficult by the often high angular zone
of acceptance by the lens system.
Similar in some respects to the reading of bar codes is
another use of the invention. In this case, however,
bar or grid locations are read by imaging through a sur-
face to determine the quality of the surface.
~or example, consider fig. 13 which illustrates the imag-
incJ of a u~iform grid of long vertical bars 400 of spac-
ing ~ (out of plane of paper) by lens 401 onto linear
- 26 -
diode array 402. Linear back illumination source 405
provides diffuse light.
In this case however, the light path includes the part
surface 410 as a mirror, whose properties are of interest
either for discrete defects such as dings and dents, or
surface finish such as microfinish or paint quality.
Object distance of lens 401 is Ll + L2. In this latter
case the contrast of the bars gives the desired answer,
and generally an alternative analo~ circuit for ~-A
conversion and amplitude analysis of the grid image pro-
vides the answer.
Consider now the case of dings and dents on sheet metal
body panels such as the car door shown, moving on conveyor
420 out of the last stage of an automatic transfer press
line.
~ere the problem is to find localized deEects on the panel
such as dings and dents caused primarily by dirt in the
press die. There are typically 002"-015" deep or raised
approxi~ately 050"-100" in diameter. This type of flaw
is different than those on the cam lobe in fig. 9 since
there is no obvious contrast difference unlike a blac~
porosity pit on a shiny cam lobe for example. ITowever
there is an appreciable localized slope change on the
part surface and this technique finds it.
The diode array scans very fast (typically 1000/sec.)
and looks for localized distortions in tne grid pattern
due to dings or dents (see inset). Alternatively, a less
desirable but workabie way is to look for localized
contrast drops.
- 27 -
Once identified the local contour distortion points and
~their magnitudë (ie. degree of distortion) can-be read
into a storage memory and used to present a defect map
of the surface. Alternatively the prese~ce of a defect
can be used to trip an ink marker to mark it 2nd/or a
red light to reject the panel.
The above disclosed circuit 440 serves to help identify
the edge locations of the bars, from which sequential
subtractions to determine local spacing changes can eas- ~
ily be made in hardware or the microcomputer 450 to de-
termine localized defects.
In a more sophisticated generally off-line processing
mode, the microcomputer can compare the variation in image
spacing tsuch as sl to sl + ~)over the whole panel
(rather than locally) and thence get overall panel con-
tour. This wor~s well on ~uasi flat surfaces. On more
curved surfaces the direct projection triangulation
technique such as figure 12 is useful, in this case pro-
jecting one or more lines whose edge locations are
monitored rather than a spot as in fig. 12 whose image
center is determined. In this case light source 405 is a
projection lamp and an auxilliary lens 451 (dotted lines)
is used to image the grid onto the surface. In this case
the object distance of lens 401 is Ll.
The use of such projected grids has a considerable use
on feature analysis and pattern recognition of objects
where it lends a shape function more or less independent
of part reflectivity. In this application, the disclosed
electronics are a definite advantage as it allcws edge
location to be determined over a wide range of light
levels and out of focus conditions (due to excessive
- 28 -
s
object depth for the lens system used). This arrange-
ment does not however have the amplication provided by
the lever arm effect of the mirror formed by the surface
(which of course can only be used when the surface is
suitably reflective at the wavelength of light used).
Note that in theembodiment shown, the raw metal surface
of the panel is seldom reflective enough by itself.and
further has random distribution of drawing compcund, die
oil, and water on-its surface which cause distortion of
the grid oreven destruction of the image. Accordingly,
a part.brushing and oil station is provided with an op-
tional air blow as well.
As panels come from the press, they are first sprayedwith a light oil mist by sprayer 480, typically the
same oil used in the die, which mixes with the wash
water. Then they are brushed with a higher quality
non shed brush, 4~2 oriented at an oblique angle such
that combined with the conveyor speed, the brush marks
are transverse to the direction of motion,-i.e. in the
direction of array scan which makes undulation of the
oil film in this direction notpickedup as desired.
This surface preparation works surprisingly well. Addi-
tionally an air blow can be provided to smooth the oil
film after the brush, but this is generally not neces-
sary.
~n alternative analysis circuit of use for the above
embodiment is to employ a phase lock loop. In this case,
the diode array grid image output is filtered by a low
pass ~ilter 46~. A phase lock loop integrated circuit
470 such as an NC 565, "locks on" to the grid image
fre~uency (l/Sl) and a band pass filter 471 provides an
output signal on a meter or similar device 472 whose mag-
nitude is roughly proportional to the magnitude of the
- 29 -
~defect.
Use of the phase lock circuit requires a prescan of
grid bars to establish the lock in, thus one must scan
somewhat more than the surface area under investigation.
In a practical example of this embodiment of the inven-
tion, a Reticon 1728 element linear array, with a grating -~~
12" long having equal spaced or 2" high and spaced 1/8"
wide. Part travel rate was 12"/sec. and a scan rate
of 1000/sec. was used. Due to the amplification pro-
vided by the lever arm effect of the surface "mirror",
this rate easily allowed more than one scan through every
flaw of interest.
It is further noted that a matrix array or other -two
dimensional scan means may be utilized to look at succes-
sive bar positions in the vertical direction thereby
not requiring the part to be moved in order to look at
successive sections of its surface (although this is
advantageously-provided "for free" in the conveyor shown).
In addition the grid may be viewed from a direction
transverse to that shown, i.e. with the lens axis trans-
verse to the part motion direction. In this case, how-
ever, a two dimensional array scan is re~uired to scan
across the surface.
It should be noted that this panel inspection technique
also works with Vidicon cam~ras, though less desirably.
Errors in said Vidicon across its field do not enter as
only local deviations are looked for. Also the angle
of incidence to the part surface is typically 5-25.
For phase locked loop operation the filter 460 is typical-
ly run at close to the clock frequency just to remove
same.
- 30 -
In the above embodiments taking first or second deriva-
-tive d~ta to describe various light pattern points has
been described. It is also of interest to take both
Eirst and second derivatives to further define li~ht
pattern positions. This need for more data is parti
cularly true in certain projected txiangulation grid
~ata as discussed relative to figure 9 and 13, where ~he
utmost resolution is desired and part reflective charac-
teristics make some data not reliable.
Another point of interest in looking at contour data
such as the varience of the grid image spacing ~1 to Sl
~ sl in figure 13b and 13c i5 that a constant frequency
filter as is desirable in other embodiments ~e.g. fig.
2B) is not desirable here as the array output frequency
is changing due to the grid image spacing thus a variable
frequency filter may here be desirable, to smooth the
discrete array elements without consideration of the
signal frequency - partic ~arly if the number of resolu-
tion elements of the array per grid image line is rela-
'tively few.
Relative to figures 2 and 5, for normal use, on object
images and diffraction patterns, it is desirable to use
a Bessel filter of at least 3 poles, with for example,
100 KHz cut off of the low pass filter for a one ~
array clock rate. Use of this filter not only filters
the array clock noise but also provides a desirable
d~gree of averaging over array elements to eliminate
inter-element sensitivity differences.
For spot triangulation, a 6 pole or greater filter is
generally desirable, operating for example at 10 KHz
for a 1 ~lz clock rate. This provides the essential
smoothing toeliminate the effect of speckle in laser
spo-t images.
- 31 -
