Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 174 (2017) 68 – 73

2016 Global Congress on Manufacturing and Management

A Study on On-Line Mathematical Model to Control of Bead

Width for Arc Welding Process
Joon Sik Son1, Jong-Pyo Lee2, Min-Ho Park2, Byeong-Ju Jin2, Tae-Jong Yun2 and Ill-
Soo Kim2,
1
Research Institute of Medium& Small Shipbuilding, 1703-8 Yongang-ri, Samho-eup, Yeongam, Jeonnam, 526-897, South Korea.
2
Department of Mechanical Engineering, Mokpo National University, 1666 Youngsan-ro, Chungkye-myun, Muan-gun, Jeonnam, 534-
729, South Korea

Abstract

Recently, not only robotic welders have replaced human welders in many welding applications, but also reasonable seam
tracking systems are commercially available. However, fully adequate process control systems have not been developed due
to a lack of reliable sensors and mathematical models that correlate welding parameters to the bead geometry for the automated
welding process. In this paper, two on-line empirical models using multiple regression analysis are proposed in order to be
applicable for the prediction of bead width. For development of the proposed models, an attempt has been made to apply for
a several methods. For the more accurate prediction, the prediction variables are first used to the surface temperatures measured
using infrared thermometers with the welding parameters (welding current, arc voltage) because the surface temperature are
strongly related to the formation of the bead geometry. And the developed models are applied to prediction of bead width as
welding quality.
© 2017
© 2016TheTheAuthors.
Authors. Published
Published by Elsevier
by Elsevier Ltd. Ltd.
This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the organizing committee of GCMM2016.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of the 13th Global Congress on Manufacturing and Management


Keywords: On-line empirical model, Optimization, Bead height, Robotic arc welding, Welding quality

1. Introduction

Controlling the arc welding process requires a closed loop approach whereby sensed information is interpreted,
and alteration is made to the process where necessary. Firstly, the system must receive the desired output
parameters and then select the necessary parameters based on these output parameters, next it must initiate welding
and finally monitor the welding outputs to determine the success of the weld. Subsequently, based on this process
monitoring, the mathematical model must be able to take positive action for any corrections needed. Most
conventional real-time, closed loop controlled welding systems attempt to control major output parameters such
as bead width. Hunter et al. [1] not only employed logarithmic equations to model the GMA welding parameters,
but also adapted capacitive distance transducers for gun positioning and ultrasonic sensing to monitor wire stick
out. Richardson et al. [2] have been employed an optical sensor coaxially mounted with a GTA welding electrode
to provide a pattern of reflected arc light for a computer algorithm to interpret joint width and location, and give
successful joint tracking. Other closed loop system has been developed by Smartt et al. [3] for real time control
of reinforcement area and cooling rate. Doumanidis et al. [4] have attempted to derive simple dynamic models in


* Corresponding Ill-Soo Kim. Tel.:+82-061-454-3455; fax: +82-061-452-6376.
E-mail address: ilsookim@mokpo.ac.kr

1877-7058 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of the 13th Global Congress on Manufacturing and Management
doi:10.1016/j.proeng.2017.01.146
 Joon Sik Son et al. / Procedia Engineering 174 (2017) 68 – 73 69

their attempt to control bead width, bead penetration, reinforcement area, heat affected zone width and rate of
cooling at the center line of the weld. Mathematical based control systems are not suited to truly real-time adaptive
control because of their inability to isolate the welding parameters, thereby resulting in excessive processing times.
These systems are not capable of taking account of heuristic's and the relative effect of each welding parameter.
Mathematical systems are also inflexible to changes in hardware due to their rigid control laws and consequently
are unable to learn from successes and errors. However, it can only consider fuzzy logic or expert systems that
are capable of a learning function [5].

This paper presents an on-line mathematical model for predicting bead width and for investigating the effects
of various process parameters using the empirical models (on-line interaction and on-line quadratic models).
Based on the experimental results, the on-line multiple regression models for predicting bead width are established.
These two kinds of models are verified by data obtained from additional multi-pass butt welds, and compared.
Finally predictive behaviors and advantages of each model are discussed.

2. Experimental Procedures

In this study, the bead geometry, an important role in determining the optimal welding conditions, is employed
to study the welding quality. Therefore in this study, Contact Tip to Work Distance (CTWD), gas flow rate,
welding speed, arc current, welding voltage were chosen to investigate the effects of process parameters and
develop the empirical models for predicting top-bead width as output parameter. Statistically designed
experiments that are based upon factorial techniques, reduce costs and provide the required information about the
main and interaction effects on the response factors. All other parameters except these were fixed. The
experimental data that included five process parameters on top-bead width at 2 levels were obtained by using a
welding robot. The design matrix that has 32 experimental weld runs where each row corresponds to one
experimental run with two replications. In this study, the mean of these replications was considered output
parameters to utilize the development of empirical models. The 150 x 200 x 4.5mm3 SS400 materials and steel
wire with a diameter of 1.2mm were employed for the experiment. Data collection and evaluation has been carried
out using the robot welding facility. The specimen was cut transversely from the middle position using a wire-
cutting machine. The selection of the electrode wire should be based principally upon matching the mechanical
properties and physical characteristics of the base metal. Secondary consideration should be given to items such
as the equipment to be used, the weld size and existing electrode inventory. 1.2‫׎‬flux-cored wire diameters and
100% CO2 shielding gas was employed in experiment. The block diagram for experimental setup in order to
develop the on-line mathematical model for bead width is shown in Fig. 1. The mechanical properties and
chemical compositions of base metal and electrode wire are shown in Tables 1~2.

Fig. 1 Block diagram for experimental setup

Table 1 Mechanical properties of base metal

70 Joon Sik Son et al. / Procedia Engineering 174 (2017) 68 – 73

Tensile Yield Impact Hardness

Material Elongation(%)
strength(MPa) point(MPa) value(MPa) (Hv)
Base metal
400 250 23 60 160
SS400
Filler wire
580 520 29 70 -
KT-71(Kiswel co.)

Table 2 Chemical composition of base metal

Element(%)
C Si Mn P S Cu Cr Ni Fe
Material

Base metal
0.15 0.05 0.697 0.013 0.007 0.041 0.087 0.503 Bal.
SS400

Filler wire
0.04 0.55 1.25 0.015 0.012 - - - -
KT-71(Kiswel)

3. Development of On-line Empirical Model

The experimental results have shown that bead width is influenced by welding parameters such as welding
speed, welding current and arc voltage. It was therefore approved that an approach for procedure optimization
could successfully carry out development of on-line empirical model to find optimal welding parameters which
would produce welds of a given quality standard. With six parameters, the response parameter Y could be and of
bead width dimensions under considerations and express as follows:

Y = f ( S ,V ,C ,T1 ,T2 ,T3 ) (1)

Where, Y is the measured bead width in [mm], S is the welding speed, V is the arc voltage, C is the welding
current and T1, T2 and T3, are surface temperatures of weld specimen.

3.1. On-line interaction model

To develop the on-line interaction model, all chosen welding parameters and interaction factors are given below:

Y k0  k1S  k2V  k3C  k4T1  k5T2  k6T3  k12SV  k12SC  k14ST1

 k15ST2  k16ST3  k23VC  k24VT1  k25VT2  k26VT3  k34CT1
 k35CT2  k36CT3  k45T1T2  k46T1T3  k56T2T  k123SVC  k124SVT1
 k125SVT2  k126SVT3  k134SCT1  k135SCT2  k136SCT3  k145ST1T2
 k146ST1T3  k156ST2T3  k234VCT1  k235VCT2  k236VCT3  k345CT1T2
 k346CT1T3  k456T1T2T  k346CT1T3  k456T1T2T
(2)

The following on-line interaction model for bead width was developed and presented as following:
 Joon Sik Son et al. / Procedia Engineering 174 (2017) 68 – 73 71

WI 40.639  1.469S  0.735V  0.051C  0.006T1  0.002T2  0.031T3

 4.169u 10 4 SC  3.391u 105 SVT1  5.9 u 105 SVT2  1.8 u 105 SVT3
 8.970u 107 ST2T3  3.3 u 107VCT1  1.8 u 106VCT2  8.8 u 108VT1T3
 4.474u 107VT2T3  2.264u 108 CT1T3  1.8 u 109 T1T2T (3)

Comparison between the predicted and measured bead width using on-line interaction model indicates in Fig.
2. The on-line interaction model for bead width was shown good performance. Fig. 3 represents the error of
predicting results of bead width by calculated from on-line interaction model. Table 1 represents performance of
on-line interaction models for predicting bead width. As shown in Table 3, bead width was predicted very
accurately more over 99% in PAM.

30
Error of the predicted bead width

25

20
The number of errors

15

10

0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Error [mm]

Fig. 2 Comparison between the predicted and measured bead width using on line interaction model

16
Bead width using W I

14
Measured bead width [mm]

12

10

4
4 6 8 10 12 14 16
Predicted bead width [mm]

Fig. 3 The error of the predicted bead width using on-line interaction model

Table 3 Performance of on-line interaction model for prediction of the bead width

Performance Bead width( WI )

PAM (%) 99.4186

Standard deviation 0.3372

4. Result and Discussion

To select the most accurate model, additional experiments were carried out. The values of the three welding
parameters were chosen for the additional experimental runs. Specially, arc voltage and welding current were
72 Joon Sik Son et al. / Procedia Engineering 174 (2017) 68 – 73

changed during GMA welding process in order to survey the reflation of the developed models according to
change the welding condition. The measured position of infrared thermometer was same as the original
experiments. Other experiment conditions have been fixed. The welding conditions including CTWD, gas flow
rate, welding speed, arc current and welding voltage were employed as the input parameters. The output parameter
is the bead width calculated by each model and the corresponding errors of prediction. From the results calculated
from the developed on-line models based on additional experiments for this study, it is evident from the two on-
line models developed that reasonable agreement between experimental and calculated bead width is shown even
when the scatter about the calculated results using two empirical equations (on-line interaction and on-line
quadratic models) is considerable. Based on the on-line quadratic model developed that determine a given bead
width and provide useful guidelines for systems which control bead width, a limited range of welding conditions,
the effect of each process parameters and their significant interactions on bead width were computed and plotted.
To compare the performance of on-line empirical model, comparison with the measured results of bead width has
been made. The performance evaluation of developed on-line model has been completed for the bead width in
trials No. 1 and No. Fig. 4 shows the comparison of the measured and predicted bead width through the developed
on-line quadratic model. As shown in Fig. 4, the values of predicted bead width are much closed to the values of
the measured bead width. Fig. 5 shows the error of predicted bead width using on-line learning neural network
model. In addition, the error of the predicted bead width using the developed on-line quadratic model is
concentrated immensely in lower values. Therefore, it is verified that the performance of the developed on-line
quadratic model is higher than that of the developed on-line interaction model.

12 12
Measured bead width Measured bead width
11.5 Predicted bead width 11.5 Predicted bead width

11 11

10.5 10.5
Bead width [mm]

Bead width [mm]

10 10

9.5 9.5

9 9

8.5 8.5

8 8

7.5 7.5

7 7
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Weld length [mm] Weld length [mm]

(a) Trial No. 1 (b) Trial No. 2

Fig. 4 Comparison of the measured and predicted bead width using on-line empirical model

2 2

1.8 1.8

1.6 1.6

1.4 1.4

1.2 1.2
Error [mm]
Error [mm]

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Weld length [mm] Weld length [mm]

(a) Trial No. 1 (b) Trial No. 2

Fig. 5 The error of the predicted bead width using on-line empirical model

For further verification, the comparison of the measured and predicted bead width using the developed on-line
quadratic model has been performed and represented in Table 2 with PAM, standard deviation and average error.
The developed model has predicted very accurately. In the comparison of standard deviation and average error,
the predicted bead width showed the most concentrated distribution. Eventually, it was concluded that the
developed on-line model has a predictive ability that is superior to the other models.
 Joon Sik Son et al. / Procedia Engineering 174 (2017) 68 – 73 73

Table 4 Performance of on-line empirical model for prediction of the bead width

Trial No. 1 Trial No. 2

PAM(%) 100 100

Bead
Standard deviation 0.2057 0.1349
Width
Average error 0.1617 0.1064

5. Conclusions

The one-line empirical model to predict optimal welding parameters on the required weld width and to
investigate the effects of welding parameters on the bead width for the GMA welding process has been developed.
Empirical equations (on-line interaction and on-line quadratic models) for on-line controls can find the
interrelationship between welding parameters and bead width for the robotic GMA welding process. Arc voltage
shows no significant effect on the bead dimension. The comparison with values of coefficient of multiple
correlations for on-line interaction and on-line quadratic equations present no differences, which indicates that all
equations are reasonably suitable. The developed two on-line empirical models are able to predict the optimal
welding parameters required to achieve desired bead width and weld criteria, help the development of automatic
control system and expert system and establish guidelines and criteria for the most effective joint design. A rule-
based expert system can be incorporated with the developed system to integrate an optimized system for the
robotic GMA welding process. It has been realized that with the use of the developed system, the prediction of
bead width becomes much simpler to even a novice user who has no prior knowledge of the robotic GMA welding
process and optimization techniques.

Acknowledgements

This research was financially supported by Jeollanam-do of technology development project for Ppuri Industry.

References

[1] J. J. Hunter, G .W. Bryce, and J. Doherty, “On-line control of the arc welding process”, in Proceedings
of the 2nd International Conference on Computer Technology in Welding, Cambridge, UK, 1988, pp.
1-12.
[2] R. W. Richardson, A. Gutow, R. A. Anderson, & D. F. Farson, “Coaxial weld pool viewing for process
monitoring and control”, Welding Journal, Vol. 63, no.3, pp. 43-50, 1984.
[3] H. B. Smartt, P. Einerson, A.D. Watkins, and R.A. Morris, “Gas metal arc welding process sensing and
control, Advances in Welding Science and Technology”, in Proceedings of an International Conference
on Trends in Welding Research, Gatlinburg, Tennessee, USA, 18~22, May, 1986, pp. 461-465.
[4] G. Doumanidis, M. Hale, and D. E. Hardt, “Multivariable control of arc welding processes, Advances
in Welding Science and Technology”, in Proceedings of an International Conference on Trends in
Welding Research, Gatlinburg, Tennessee, USA, 1986, pp. 449-457.
[5] H. B. Smartt, “Arc-welding process”, in Welding: Theory and Practice, Elsevier Science Publishers,
1990, pp.175-208.