Note: Descriptions are shown in the official language in which they were submitted.
2004311.
,
ELPATRONIC AG, Baarerstrasse 117, 6300 Zug, Switzerland
Case 88-122
Switzerland
ARRANGEMEN~ FOR MONITORING THE QUALITY OF ELECTRIC WELDS
The invention relates to an arrangement for
monitoring the quality of electric welds on bodies for the
production of cans, having at least one device for
detecting at least one welding parameter profile, having a
store for storing the first welding parameter profile and
having a comparator for comparing the first welding
parameter profile with a second parameter profile measured
later, in order to determine their degree of agreement.
In a known arrangement of this kind (~P-A2-0 153
298), the aim is to detect precisely the variation of the
welding parameter in time and to define precisely a desired
run of data associated with a welding operation, in the
course of which various marginal conditions, particularly
the kind of material, the th'ickness of material, a coating
of the material, the basic parameter of the particular
welding machine etc., are to be taken into consideration
and the data obtained are to be combined and evaluated in
accordance with various programs. For this purpose, in the
known arrangement, a favourable course of the welding
parameter is determined for each individual welding
position on a sample workpiece having a plurality of
3~
2004311.
welding positions, and is stored. The store can be
programmed through a microprocessor so that it is possible
to select a welding pattern which is recognized as
particularly favourable and to store its data profile as a
desired profile for following like welding operations.
~hen, when the welding machine is in series operation, the
associated reference data for the particular welding
position, which are retained as a desired profile in the
store, are recalled. As a result of interrogating the
instantaneous values of the welding parameter during the
welding operation, a data profile results which is
compared, in a comparator, with the stored data profile
corresponding to the desired course of the welding
operation. Output signals are generated according to the
degree of agreement found. If a minimum degree of
agreement is found, the weld can be evaluated as "good".
It is not stated, however, how the limiting value for the
minimum degree of agreemen~ between desired and actual
course, for which the weld can be evaluated as "good" is
set. It is to be assumed that a fixed value is set which
is at a certain margin from the desired course. In order
that the set limiting value may be able to be used for the
whole length of the weld seam, without too many
satisfactory welds being rejected as "poor", a relatively
large set limiting value must be selected. Otherwise great
deviations usually occurring at the beginning and end of
the weld seam would continuously lead to erroneous rejects.
The known device is therefore not very sensitive and not
200~311
very precise in quality control.
It is the ob]ect of the invention to increase the
sensitivity and the precision in quality control in an
arrangement of the kind mentioned at the beginning.
According to the invention, this problem is solved
in that an evaluation device is provided for the statistical
evaluation of mesured welding parameter values of a number
of can bodies found to be good, by determining an average
welding parameter profile which is stored in the store as
the first welding parameter profile, and by determining a
reject limiting-value band enveloping the average welding
parameter profile from the mean value of the welding
parameter values measured for the first welding parameter
profile, and from the product of the standard deviation of
these welding parameter values assumed to be normally
distributed and at least one sensitivity factor, and that,
for can bodies subsequently produced, the comparator
determines whether the second welding parameter profile
measured for these lies within the reject limiting value
band.
According to the present invention there is
provided an arrangement for monitoring the quality of
longitudinal electric welds on bodies for the production of
cans, having at least one device for detemrining at least
one first weld parameter profile (F), having a memory to
store the first weld parameter profile, and having a
comparator to compare the first weld parameter profile with
a second weld parameter profile measured later, in order to
determine their degree of agreement, characterised in that
the said device contains an evaluating device for the
statistical evaluation of measured weld parameter values of
profiles (F) extending the length of the weld, or can body,
the evaluating device being provided:
2004311
3a
- with a first calculating device which is set up:
- to determine an average weld parameter profile
(Fm) from a number of can bodies found to be good,
- to determine the standard deviations ai of the
weld parameter values belonging to points i of the
average weld parameter profile tFm), and
- to determine the second weld parameter profile for
can bodies subsequently produced,
- with a memory for the said average weld parameter
profile (Fm) and the associated standard deviations ~i'
- with a second calculating device for determining a
reject limiting-value band (Fg+) enveloping the average
weld parameter profile (Fm) and constituted as the
product of the means values (Fmi) of the weld parameter
values (Fm) resulting from the average weld parameter
profile (Fm)and of a standard deviation ai multiplied
by a sensitivity factor,
and in that in addition the said comparator is set up to
ascertain, in respect of the can bodies subsequently
produced, whether the second weld parameter profile
determined for these lies within the reject limiting-value
band (Fg+)~
According to the present invention there is also
provided an apparatus for monitoring the quality of electric
welds on can bodies, each can body having a weld seam formed
by a plurality of welds extending continuously along the
length of a can body each weld being associated with a
welding parameter of a given value, said apparatus
comprising:
- means for receiving weld signals indicative of
the value of the welding parameter as measured for each
weld, with a set of said parameter values for each weld
seam;
\ -
20~311
3b
- at least one means for determining from said
weld signals an average welding parameter profile (Fm)
computed from a group of acceptable weld seam parameter
value sets measured on a plurality of acceptable welded can
bodies;
- store means to store the average welding
parameter profile;
- a comparator to compare the average welding
parameter profile with welding parameter profiles measured
later on subsequently welded can bodies for determining a
degree of agreement between the average welding parameter
profile and each of said subsequently welded can body
welding parameter profiles; and
- evaluation means for providing statistical
evaluation of said subsequently welded can body welding
parameter profiles and determining a reject limiting-value
band (Fg+) enveloping the average welding parameter profile
(Fm) from computed mean welding parameter values for each
weld in said group of acceptable weld seam parameter value
sets and from the product of a standard deviation of said
welding parameter values for each weld in said group of
acceptable weld seam parameter value sets assumed to be
normally distributed and at least one weld quality
sensitivity factor;
- the comparator further generating output signals
indicative of whether each welding parameter value of said
subsequently welded can body welding parameter profiles lie
within the reject limiting-value band (Fg+)~
According to the invention, therefore, a desired
data profile of the absolute value of the welding parameter
or of the variations in the welding parameter is not found
on a sample can body but, in a learning phase, the average
value of the welding parameter over the length of the can
20~4311
body is determined from a plurality of can bodies found to
be good and, as a result of the deviation in the measured
values observed in the course of this, the two reject
20043~.
-
limiting value curves defining the reject limiting valueband are determined, while a certain sensitivity is
selected, for example in such a ~nn~r that only every ten
thousandth good can body is rejected.
In the arrangement according to ~he invention,
therefore, the measured welding parameters are ex~r;ned
with regard to their statistical behaviour. Only "good"
can bodies are included in the statistics. After the
welding machine has been adjusted so that the can bodies
produced have the desired quality, a first statistic is
built up in the learning phase. The cans produced during
this learning phase are checked at least roughly for
quality (for example by visual checking, micrographs,
tearing-open tests, X-ray ~x~ ination etc.). Experiments
have shown that the welding parameter values measured for
the individual welding positions are normally distributed,
that is to say they have Gaussian distribution or normal
distribution. This means that the statistical
characteristics of the welding parameter values
(distribution curve) are determined by average value and
variance.
The important difference from the known
arrangement is therefore to be seen in that by determining
a reject limiting-value band enveloping the average welding
parameter profile from a plurality of can bodies found to
be good and selective probability for the rejection of a
can body which is actually good, a better sensitivity and
an optimum precision are achieved with the arrangement
~004311
s
according to the invention. For this purpose, the user
only has to fix the sensitivity in the arrangement
according to the invention. In the usual case, the
sensitivity for exceeding the lower and upper limiting
value respectively can be set separately. The criterion
for this is preferably the number of erroneous re~ects per
million of weld spots. The question as to how many
erroneous rejects are permissible can be answered with
reference to the quality requirements in each case.
According to the present invention an optimwm tolera~ce barld
is determined for each specific production condition, kind
of sheet metal, machine setting or the like. In contrast
to the known arrangement where two straight lines are
simply used and a constant upper and lower limiting value
over the whole length of the can body is preset by a
potentiometer, in the arrangement according to the
invention a tolerance profile is used which is adapted to
the production situation. This tolerance profile makes it
possible to work with a broader reject limiting-value band
in the beginning and in the end region of the length of the
can body, where experience has shown that the deviations in
the measured welding parameter values are particularly
great, than over the middle of the length of the can body,
for example, where experience has shown that the deviation
is considerably less.
By means of the statistics used as an aid in the
arrangement according to the invention, an automatic
~,. i .
Z00~3~
adaption to the constant quality of the tested can bodies
can be achieved. Even if, for example poor sheet-metal
material or "unsteady" machines cause widely scattered
measured values, this is automatically detected by the
higher deviation value and the limiting values become
greater. What the arrangement according to the invention
does here automatically, is often done unconsciously by a
person setting a limiting value manually since they will
adJust the potentiometer so that a good can body is
rejected sporadically but not too often.
After the first statistics of the good can bodies
have been made available by the learning phase and so the
reject limiting values have been fixed, the can bodies
which have been found good continue to be followed
statistically in the above-mentioned manner by the
arrangement according to the invention during the whole
produc~ion. As a result, it is possible to produce a
quality display.
Advantageous developments of the invention form
the subject of the sub-claims.
Examples of embodiment of the invention are
described in more detail below with reference to the
drawings.
Figure 1 shows diagrammatically, the construction
of the arrangement according to the invention for
monitoring the quality of electric welds on bodies for the
production of cans,
Figure 2 shows, as a detail, a block circuit
X004311.
diagram of a statistical evaluation device according to
Figure 1,
Figure 3 and 4 are explanatory diagrams, and
Figure Sa - 5d are diagrams to explain a
continuous quality control which can be carried out by
means of the arrangement according to the invention.
An arrangement for monitoring the quality of
electric welds on bodies 10 for the production of cans is
described below in connection with an alternating-current
seam welding apparatus which latter, according to the
diagrammatic partial illustration in Figure 1, comprises a
welding arm 11 on which the can bodies 10 are moved
forwards (towards the right in Figure 1) and in the course
of this are welded together in the region of their
Longitudinal seam between a lower welding roller 13 secured
to the welding arm 11 and an upper welding roller 12
pressed against the lower roller by means of a spring 14,
as a result of a welding current flowing between the
electrode rollers through the can-body material. Each
half-wave of the welding current leads to the production of
a welding spot. In successive half-waves of the welding
current, the welding current has the one polarity or the
other. With regard to further details, reference should be
made to the company brochure of Soudronic AG, "Warum
SOUDRONIC Frequenzwandler-Schweissmaschinen", September
1979, for example pages 6 - 8.
In the machine, for example at the upper
electrode roller 12, according to the diagrammatic
- Z0043~.
illustration in Figure 1, one (or more) welding parameter
Fi is measured ~y sensors (not illustrated). The welding
parameter Fi may be the welding current, the voltage at the
wel~g position, the product of these quantities
(power/energy), the seam temperature, t~e travel of a
pendulum roller head which carries the upper electrode
roller 12 or the like. Each welding parameter Fi is
preprocessed in a data processing device 20 in such a
manner that a digital numerical value is produced for each
welding spot i (with i = 1, 2, 3,...n). Such a
preprocessing may consist in that the electrical signals of
the sensors are filtered for the purpose of eliminating
superimposed interference signals. Another method of
processing the measured values may consist in forming the
said product of welding current and voltage and integrating
it over the duration of a welding spot (spot energy).
These preprocessed measured values are meant when welding
parameters or welding parameter values or spot values are
referred to below. In the arrangement described below, the
spot values supplied by the individual sensors and
processed in the manner described above are examined with
regard to their statistical behaviour for each criterion
(energy, temperature etc.).
For this purpose, the welding parameter values
are supplied to a statistical evaluation device 22.
According to the more detailed illustration in Figure 2,
the statistical evaluation device 22 contains a first
calculating device 22.1, to the input of which the welding
;~004~11.
-
parameter value Fi is fed and the output of which is~g~ne~ted to the input of a store 24. The output of the
store 24 is connected to one input of a second calculating
device 22.2. This has a further input which is connected
to the output of an input device 22.3 which comprises a
search table. The search table contains a number of
sensitivity factors which can be addressed through the
inputs of the input device 22.3 according to the
sensitivities E1 and E2 selected. The selectivity of the
sensitivities El, E2 is represented symbolically by
potentiometers 22.4, 22.5. The search table may also be
replaced by a calculation by means of microprocessors (not
illustrated). The second calculating device 22.2 has two
outputs which lead to two inputs of a comparator 26 (Figure
1). At a third input, the comparator 26 receives the
welding parameter values Fi. The output of the comparator
26 is connected to a data bus 27 to which parallel signal
processing devices 20, 32, 34, 36, 38, 40 are connected,
which are explained in more detail below.
First the first calculating device 22.1
determines an average welding parameter profile F from the
welding parameter values Fi, which profile is stored in the
store 24 and is illustrated in the form of two different
curves in Figures 3 and 4.
There are two possibilities for determining the
curves Fm, namely as the arithmetical mean value of the
absolute measured values or as the arithmetical mean value
of the difference formed between each two successive
200431~.
-
10measured values.
In a learning phase, for which 20 to 100 can
bodies are generally used, a first statistic is built up.
On each can body, the ith spot is measured and the welding
parameter values found in all ith 5pO~S of all the can
bodies are averaged, which produces mean values Fmi over
the length 1 of the can bodies from the first to the nt
spot (Fig. 3 and 4).
The welding parameter values have a normal or
Gaussian distribution. The distribution function is
characterised by the mean value Fm and the standard
deviation a. A Gaussian curve is clearly defined by the
mean value Fm and the standard deviation a The standard
deviation is a measure of the extent to which the measured
values are scattered and defines the range in which 66% of
all measured values lie. The two base points of the
standard deviation (one base point in the negative
direction at one side of the mean value and one base point
in the positive direction on the other side of the mean
value) and the mean value thus clearly define a Gaussian
curve. With regard to further details, reference should be
made to the hand~ook "Statistische Methoden und ihre
Anwendungen" by E Kreyszig, Verlag Vandenhoeck & Ruprecht,
Gottingen, second, unaltered reprint of the 7th edition,
pages 125-129. The first calculating device 22.1
calculates Fmi and ai for each spot i consecutively. The
calculation of Fmi is effected as an arithmetical mean
value as explained above. The calculation of ~i is
2004311.
effected in accordance with the equation
I m
ai = ~ l(Fii mi) (1)
in which
ai = st~n~rd deviation in the i spot
m = number of can bodies in the learning phase
i = current can body
Fij = random sample value of the it spot of the
j can body
m
mi m J-l ij
The mean value profile F and the profile of the
standard deviations ~i are stored in the store 24 from
which they can be fetched when necessary by the second
calculating device 22.2. The second calculating device
22.2 calculates the reject limiting-value band F + from the
following equations
F . = F . + z a. (2)
gl~ ml
gi- Fmi Z2 ~i (3)
with i = 1, 2, 3...n and
Fgi = reject limiting value in the i spot of
the welding seam,
Fmi = mean value of the welding parameter in the
it spot of the welding seam,
Zl = sensitivity factor or reciprocal quantity
of the sensitivity for upper reject
limiting value, and
- ZO(~43~1.
12
Z2 = sensitivity factor or reciprocal quantity
of the sensitivity for lower reject
limiting value,
ai = st~n~Ard deviation for the ith spot of the
welding seam,
In accordance with the characteristics of normal
distribution, there is a definite probability that the
limiting value Fgi+ or Fgi will be exceeded. According to
the magnitude of Z~i' can bodies which are actually good
are rejected more or less frequently. The smaller Z~i is,
the more sensitively does the arrangement react to defects
but good can bodies are also rejected more frequently.
In the following explanations, a reject limiting-
value band is assumed, for the sake of simplification,
which is symmetrical to the average welding parameter
profile F , that is to say the case El = E2 = E and
Zl Z2
The "good/bad" criterion is derived from the
desired sensitivity of the arrangement. This is defined so
that a proportion, determined by the user, of good can
bodies may be classified as bad. Thus an objective
rejection criterion is laid down which is here called the
sensitivity E. It is defined as the number of erroneous
rejects per million of can bodies. The sensitivity value
E = 100 therefore means 100 erroneous rejects per 1 million
of can bodies or that, on the average, every ten thousandth
can body is accidentally rejected. In a body welding
machine which produces 600 can bodies per minute, this
- Z0043~1.
13
means that a good can body is accidentally rejected every
quarter of an hour.
The sensitivity factor z can be addressed through
the potentiometers 22.4,22.5 in the search table 22.3 which
contains a number of sensitivity factors concerned. z is
the reciprocal value of the sensitivity E. The greater z
is, the less sensitive is the arrangement. Values
concerned for z are shown from the said handbook, pages 128
and 129, from which it can be seen that (with Fm written
instead of ,u):
About 95% of all values lie hetween F - 2a and
m
+ 2a,
about 99 3/4% of all values lie between F - 3
m
and Fm ~ 3a~ and
about 99.9~ of all values lie between F - 3.29
m
and Fm ~ 3.29~.
In these cases therefore
z = 2
z = 3
z = 3.29
applies.
According to the literature, for example
"Handbook of Mathematical Functions", Ed. M. Abramowitz and
I. Stegun, Dover Publications, Inc., New York, December
1972, page 933, z can be calculated with a satisfactory
approximation as follows:
Z = ~ ~ 2 ln (1 - A2)
200431~ .
-
14in which A = the probability that the limiting value will
not be exceeded.
Therefore if for example 99.0000% of all spots
should lie within the (symmetrical) re;ect limiting-value
band, A = 0.999999 and the calculation gives z = 4.54.
As a result of experience gained, a value between
3 and 5 and preferably the value 4.54 is selected for z.
The last-mentioned value corresponds to 99.9999%, that is
to say the case in which per 1 million of weld spots, one
is accidentally judged as "bad". With 100 weld spots on
the body of a medium sized can, every ten thousandth body
is thus accidentally rejected on the average.
The second calculating device 22.2 takes the
addressed value of z from the input device 22.3 and then
calculates the reject limiting-value band in accordance
with the equations (2) and (3), the associated values of
Fmi being taken from the store 24 each time. Two reject
limiting-value curves thus calculated, which define the
band at the top and bottom are designated by F + and Fg_
respectively in Figs. 3 and 4. The reject limiting-value
curves Fg~ and Fg are stored in a store not illustrated in
the second calculating device and delivered to the
comparator 26 when necessary.
The processing and evaluation procedure described
above is carried out for the 100 bodies found to be good,
in the learning phase. Subsequently, that is to say during
the production phase of the welding machine, the welding
parameter values are processed as in the learning phase but
Z0043~.
-
15are then preferably supplied to the comparator 26 which
therefore ascertains, for bodies 10 produced following on
the learning phase, whether the second welding parameter
profile measured for these which, more precisely, is
likewise formed from the corresponding mean values Fmi,
lies within the reject limiting-value band Fg+. The
comparator 26 delivers an output signal depending on the
result of the comparison to the signal processing devices
30 to 40. These include a printer 30 which prints out the
result of the comparison, a display device 32 which
displays the result of the comparison, a signal lamp 34
which indicates whether a body was "good", an indicator
lamp 36 which indicates whether a body was "bad", an
ejector 38 which ejects the body simultaneously with the
actuation of the signal lamp 36, and a controller 40 which
readjusts the measured welding parameter with a correcting
action if a deviation trend should occur, which is
explained in more detail below.
From the fact that certain faulty mechanisms
become apparent as a result of the fact that a plurality of
successive spots cause statistically significant
deviations, further (lower) limiting values can be fixed as
follows:
2 m
in which case only one defect is signalled if F2 is
exceeded on two successive measurements, F3 on three
successive measurements etc. The deviations Z~i can again
be defined in each case so that, for example, every
20043~1.
16
thousandth good body is rejected. The reject limiting-
value band F2+ for F2 is shown in Fig. 3.
If two isochronous criteria (for example seam
sque~i~g and spot energy) are available, a further
possibility for forming the limiting value results in such
a manner that a so-called correlation limiting value Fk is
defined. In this case, a rejection is effected if, for a
certain weld spot, both spot values exceed their
correlation limiting value Fk.
A simplification of the procedure results if not
every spot i is evaluated for the statistics as in Fig. 3
but regions are combined. In the example illustrated in
Fig. 4, only three statistics are given instead of n,
namely for the regions A, B and C, that is to say for the
beginning, middle and end re-spectively of the body Each
region comprises 10 to 20 weld spots which are combined to
form a mean value per region for the statistical
evaluation.
The arrangement described here is suitable for
continuous quality control which is described below with
reference to Figs. Sa,5d.
After a first statistic of the good bodies has
been made available by the learning phase and so the reject
limiting-values have been fixed, the bodies which have been
found to be good continue to be followed statistically by
the above-mentioned methods during the whole production
run. Thus it is possible to produce a quality display.
The illustration in Fig. Sa is the starting
201~43~
-
17
position with little deviation and high sensitivity,
E = 1000. Fig. 5b shows the starting position with average
deviation and low sensitivity, E = 100.
Thus the illustrations according to Figs. Sa and
5b show the mean values and the scattering of the measured
values with the selected sensitivity E = 1000 and E = 100
respectively. The scattering and the sensitivity are
relative values, the physical significance of which does
not need to be known to the user. A high sensitivity means
a higher probability that here and there a good body will
be rejected A high dispersion can have various causes:
poor material, unstable machine setting, contamination of
the machine etc.
Figs. Sc and 5d show possible variations after a
starting position according to Fig. 5a. Fig. 5c shows a
drift in the measurements and Fig. 5d shows a drifting mean
value and increasing dispersion, in which case the sensor 1
sets off an alarm. The drifting away of the mean values
might have the following causes:
- Drift of the sensors 1,2 as a result of heating
of the machine,
- Variation in a machine setting or in the
material properties.
The variation in the dispersion could have the
following causes:
- Increasing contamination,
- Something on the welding machine has become
"shaky".
20043~1.
18
As soon as an alarm value is reached (represented
graphically in Fig. 5d by the fact that the measured values
represented by plus symbols are touching the boundary), the
user must check the quality of production. If the quality
is still good, he can reach the illustration shown in Fig.
5a again ~y pressing a function key "learning". In this
case, the deviations have been caused by influences which
do not impair the welding quality (for example temperature
drift of the measuring device). This alarm value can be so
defined for example that it is equal to ten times the
value of the permitted reject quota.
The data formed by the continuous quality check
make it possible to intervene automatically with a
correcting action, for example in that the welding current
is readjusted by means of the controller 40 in such a
manner that the mean values return to the starting position
shown in Fig. Sa.
