Beruflich Dokumente
Kultur Dokumente
Abstract
This work was aimed at developing a methodology for the estimation of the contact tip-to-workpiece distance (stand off) in the gas metal arc
welding (GMAW) process. The proposed methodology was based on the measurement of the electrical resistance between the contact tip and
the workpiece during the short circuits in the dip transfer mode of metal transfer (dip resistance). The time periods in which the short-circuits
were occurring, were identified based on the positive cycles of the time derivative of the welding current signal, together with the analysis of
the welding voltage. The experimental part of this work was carried out in three basic parts: (a) determination of the working envelop for the
welding parameters; (b) acquisition of welding data from controlled welding trials and the development of multiple regression models; and
(c) validation of the chosen model. The weld beads were laid using an AWS A5.18/79 welding wire of 0.8 mm diameter, C25 shielding gas
(Ar + 25% CO2 ) on 1/4 in. (6.25 mm) mild steel plate workpieces (ABNT 1020). A multiple linear regression model was developed having as
independent variables the wire feed speed, the average welding current (measured on-line), the average dip-resistance (measured on-line) and
some second order interactions between these variables, and the predicted stand-off as the dependant variable. The standoff obtained from
the validation weld runs were in good agreement with the predicted values. Errors around 4% were observed.
© 2004 Elsevier B.V. All rights reserved.
0924-0136/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2004.09.002
4 S.C. Absi Alfaro et al. / Journal of Materials Processing Technology 157–158 (2004) 3–7
that data collection would begin when the welding current Otherwise the derivative and the short-circuit resistance are
rose above 10 A. During data collection graphs of voltage set to zero. These results are sent to a text file in the form of
and current were shown for each window. They were based a matrix.
on 1024 samples. When the welding was finished the data
capture was stopped and the following results were sent to a 2.4. Stochastic analysis
text file:
For this phase a program for statistical analysis was used.
size of the window (1024 samples per window);
With it various analyses may be undertaken until an adequate
sampling frequency (2500 Hz);
model of the process under investigation is obtained. The
set stand-off at the welding source (SO);
following steps were taken:
wire feed speed set at the welding source (WFS);
voltage set at the welding source (Vset ); establishing the variables to be used in the model;
average voltage obtained at the welding source (Vmfmt ); establishing that these variables are independent;
average current obtained at the welding source (Imfnt ); reaching an equation that may be used to estimate stand-
signals from channel 0 (voltage); off.
signals from channel 1 (current).
Forward stepwise regression was used to discover those
variables that had the greatest influence on the process. This
2.3. Analysis of the welding data
involves including and excluding variables and finding the
combination that produces the largest F-value (variance ra-
Another computer program was prepared to analyze the
tio).
data after it had been collected. This program reads the data
The independence of the variables used in the study was
that is produced in the previous step. It analyses each of
not all that clear. The requirements for the analysis are that
the data “windows” and calculates the means of the weld-
the variables are (i) independent, (ii) random, and (iii) have
ing voltage, current and resistance. These values are shown
a normal distribution of errors. Ridge regression was used to
here plotted as a function of time. Here the time for each
test the independence of the variables. This provided the “co-
window is 0.4096 s. If a regular frequency of short-circuits
efficient × theta” graph. This should and did show a tendency
can be observed in these graphs the program moves on to an-
towards zero correlation.
other phase and the first derivative of the welding current is
The next step was to discover the mathematical model to
obtained.
be used to estimate the stand off by multiple regression. The
As is well known the derivative of an increasing function
analysis of variance was used to choose the model and the
is positive and that of a decreasing function is negative. When
largest F-value was the criterion of choice.
a short-circuit occurs the welding current increases rapidly
as may be seen in Fig. 2. Further, while the short-circuit is
happening the arc practically vanishes and as a result the re- 2.5. Validation of the model
sistance to the passage of the current tends to drop quickly
with the result that the short-circuit current increases. Based The model was validated by carrying out new experiments
on this when the welding current derivative is positive the val- where the same variables, equipments, and procedures were
ues of the short circuit resistance and its mean are calculated. used as had been used in the estimation of the regression
equation. However, different values (albeit within working
limits) were used to those that had been used to obtain the
model. These tests were conducted according to a 23 factorial
design. The results furnished by the model were compared
with the set stand off values obtained before the start of the
welding process.
3. Results
Table 2
Analysis of the voltage and current signals
NRO SO (mm) WFS (m/min) Vset (V) Vmfnt (V) Vmexp (V) Imfnt (A) Imexp (A) Rccmedia (Ohm)
1 8 4 21 20 20 83 83.8 0.308384
2 8 4 23 20.2 20 85 85.3 0.290929
3 8 6 22 20 20.6 120 122.7 0.218371
4 8 6 23 21.8 21.6 121 124.2 0.222263
5 8 8 25 23.5 23.2 147 155.4 0.187302
6 8 10 26 24.5 23.8 188 190.3 0.145768
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
25 16 8 24 22.8 22.6 121 122.3 0.232269
26 16 10 24 22.7 22.3 137 137.1 0.204926
27 16 10 26 24.7 24.3 142 143.6 0.200853
28 18 4 20 19.3 19.1 66 67.5 0.377326
29 18 6 22 21.5 20.9 93 95.7 0.276575
30 18 8 24 22.8 22.6 117 119.3 0.239869
31 18 10 24 22.6 22.4 133 135.8 0.209656
SO, standoff value set within the robots movement program; WFS, wire feed speed; Vset , voltage set at the welding source; Vmfmt , mean welding voltage
obtained at the welding source; Vmexp , mean welding voltage obtain from the voltage signals of the experiment; Imfmt , mean welding current obtained at the
welding source; Imexp , mean welding current obtained from the current signals of the experiment; Rccmedia , mean short-circuit resistance of the experiment.
Table 3
Result obtained from the forward stepwise variable selection
Independent variable Coefficient Standard error t-value Significance level
Constant 23.270702 4.133501 5.6298 0.0000
WFS −8.071033 0.896089 −9.0070 0.0000
Imexp 0.401666 0.050576 7.9418 0.0000
WFS × Rccmedia 53.008483 4.158347 12.7475 0.0000
Imexp × Rccmedia −3.174132 0.295681 −10.7350 0.0000
Table 7
Result of the validation of the model
Run SO WFS Vset Vmfnt Vmexp Imfnt Imexp Rccmedia SO Error SOˆ(0.8) Error
(mm) (m/min) (V) (V) (V) (A) (A) (Ohm) estimated (%) estimated (%)
1 9 7 23 21.6 21.4 131 135.3 0.19631 9.7 7.3 9.6 6.4
2 17 7 22.5 21.3 21.3 104 107.8 0.251777 17.4 2.0 17.4 2.2
3 9 9 25 23.3 23.1 157 161.6 0.174853 9.3 3.0 9.2 2.7
4 13 9 24 22.7 22.4 137 139.6 0.197023 13.4 3.1 13.3 2.3
5 17 5 21 19.9 20 81 83.6 0.315428 16.4 3.6 16.4 3.7
6 13 5 20.5 19.4 19.4 87 90.4 0.289064 12.9 0.8 12.8 1.6
7 9 5 20 19.1 18.8 97 98.8 0.265522 9.7 7.8 9.7 7.7
8 13 7 23 21.7 21.6 117 120 0.223792 12.8 1.8 12.6 2.7
9 17 9 24.5 23.1 23 124 128.6 0.225998 17.9 5.1 18.0 5.7
Mean 3.8 3.9
Standard 2.4 2.2
deviation
SO estimated, standoff obtained from the linear model; SOˆ(0,8) estimated, standoff obtained from the non-linear model.
5. Conclusions References
The work reported here permits the following conclusions: [1] G.C. Carvalho, An Adaptive Control System for Off-line Program-
The results show that the methodology is valid and that it ming in Robotic Gas Metal Arc Welding. Cranfield, Ph.D. thesis,
allows the establishment of a more complete control of the School of Industrial and Manufacturing Science, Cranfield University,
1997.
process. [2] J. Norrish, Adv. Weld. Process, Bristol, UK, 1992.
The errors that were observed may be due to: (a) the model [3] S. Saini De floyd, Na investigation of gas metal arc welding
(in part); (b) inaccuracies in the welding data; (c) initial error sound signature for on-line quality control, Weld. J. (1998) 172s–
in positioning the robot to set the stand-off; (e) errors in the 179s.
wire feeding mechanism and (f) errors in the movement of [4] N.M. Johnson, et al., Process control of GMAW: sensing of metal
transfer mode, Weld. J. (1991) 91s–99s.
the arm of the robot. [5] Welding Handbook, 6a edição, parte 2, AWS, United States of Amer-
The linear multiple regression model is a simpler model ica, 1942.
and provides better results than the non-linear model. [6] George Box, et al., Statistics for Experimenters, An Introduction to
Design, Data Analysis and Model Building, John Wiley & Sons, NY,
USA, 1972.
Acknowledgements [7] P.E. Murray, Selecting parameters for GMAW using dimensional anal-
ysis, Weld. J. (2002) 125s–130s.
The authors would like to thanks FINEP, CNPq, RHAE, [8] J.H. Choi, J.Y. Lee, C.D. Yoo, Simulation of dynamic behavior in a
GMAW system, Weld. Res. Suppl. (2001) 239s–245s.
and FAP-DF for supporting this work.