Professional Documents
Culture Documents
DOI 10.1007/s00170-012-4437-0
ORIGINAL ARTICLE
Received: 12 April 2012 / Accepted: 30 July 2012 / Published online: 1 September 2012
© Springer-Verlag London Limited 2012
match was proposed. Two stages, i.e., global and local to the seam can be precisely reflected in the image.
match, were used in the process. The big lag between The motion mechanism is composed of two axes, i.e.,
the torch and the camera is also not considered in this moving axis and adjusting axis. They are orthogonal
method. to each other. Adjusting axis is mounted on the slide
In this paper, a novel vision-based initial weld point block of the moving axis. The torch and the camera
positioning method is proposed. It is based on the are mounted on the slide block of the adjusting axis
straight line of the seam and makes best use of the geo- through a bracket. Moving axis moves the torch and
metric relationship between the two seams at the two the camera in the welding direction, which is driven by
different stages of the welding task. Only two kinds of a direct current motor. Adjusting axis is used to adjust
information such as the position of the torch read from the position of the torch to keep it in the seam, which
the rotary encoder and the image feature of the seam is driven by a stepper motor. The rotary encoder is
are necessary. In this new method, the big lag between installed on the moving axis measuring the position of
the torch and the camera has been considered. When the torch in the welding direction. The proximity switch
the initial weld point positioning is finished, it can be is installed at a fixed place of the welding system at the
guaranteed that the torch is precisely aligned with the start position, which is used to ensure that the torch
initial weld point. moves from the original point of the welding system.
The rest of this paper is organized as follows.
Section 2 describes the configuration of the welding sys-
tem. In Section 3, the geometric relationship between 3 Analysis of the geometric relationship
the two seams at the two different stages is analyzed. between the two seams
The detailed realization of the new proposed initial
weld point positioning method including the initializa- To formulate the proposed initial point positioning
tion stage and the welding stage is given in Section 4. method clearly, the whole welding task is divided into
Section 5 illustrates the experimental results and two separate stages, i.e., the initialization stage and the
Section 6 concludes this paper. welding stage. The former stage is carried out only once
for welding the first plate. However, the latter stage
is carried out every time for welding every plate. The
2 The configuration of the welding system proposed initial point positioning method is based on
the geometric relationship of the two seams at the two
A typical welding system in the container manufacture stages. The aim of the initialization stage is to provide
is shown in Fig. 1, which consists of a camera, a welding a reference for deducing the relationship between the
torch, a motion mechanism, a rotary encoder and a position of the torch and the corresponding image fea-
proximity switch. The camera is placed ahead of the ture of the seam. The working procedures of the two
torch by a fixed distance in order to reduce the effect stages are contrary. For the former stage, the torch is
of high temperature and welding splash from the torch. manually tuned to align with the initial weld point at
Its view direction is kept the same as that of the torch. first. Then, the image feature and the parameters of the
In this way, the position of the torch with respect seam line are acquired. For the latter stage, the target
image feature of the seam is first computed based on
the geometric relationship between the current seam
Adjusting
axis and that at the former stage. Then, the torch is adjusted
Rotary
encoder
automatically to align with the initial weld point based
Moving
axis
on the target and current image features of the seam.
Welding direction As shown in Fig. 1, S1 is denoted as the seam at the
initialization stage and S2 as one of the seams at the
Torch welding stage. Four points on the two seams are defined
Oc
Ot Yc
Yt as follows: P1 is the initial weld point of S1 ; P2 is the
Proximity switch Xc
Xt Zc
P1
Zt
S2
feature point on S1 that is imaged by the camera; and
θ1 P2
L P3 and P4 on S2 are defined similarly as P1 and P2 .
θ2 P4 To facilitate the analysis of the geometric relation-
S1
P3 ship between S1 and S2 , two coordinate frames are
Weld plate established, as shown in Fig. 1. The camera frame
C is established at the optical center of the camera. Its
Fig. 1 The configuration of the welding system Z c -axis is aligned with the optical axis of the camera,
Int J Adv Manuf Technol (2013) 66:1535–1543 1537
and the directions of Xc -axis and Yc -axis are the same the position of the torch and the image feature of the
as those of the image plane. The tool frame T is estab- seam is computed similarly as follows:
lished at the tip point of the torch. Its Xt -axis and Yt - ⎧
⎪ (u4 − u0 )zc4
axis are parallel with the adjusting axis and the moving ⎪
⎪ x = − L tan θ2
⎪ t3
⎪ kx
axis, respectively. ⎨
(v4 − v0 )zc4 (5)
Suppose the image coordinates of P2 are (u2 , v2 ), ⎪ yt3 =
⎪
⎪ k
its coordinates in the frame C and the frame T are ⎪
⎪
y
⎩
(xc2 , yc2 , zc2 ) and (xt2 , yt2 , zt2 ), respectively. From the zt3 = zc4 − h
pin hole model of the camera, the image coordinates
where (u4 , v4 ) are the image coordinates of P4 , zc4 is the
of P2 are computed as follows:
z-coordinate of P4 in the frame C, (xt3 , yt3 , zt3 ) are the
⎡ ⎤ ⎡ ⎤⎡ ⎤ coordinates of P3 in the frame T, and θ2 is the angle
u2 kx 0 u0 x c2
between S2 and the Yt -axis.
z c2 ⎣ v 2 ⎦ = ⎣ 0 k y v 0 ⎦ ⎣ y c2 ⎦ (1)
The relationship between the position of the torch
1 0 0 1 z c2
and the corresponding image feature is shown in Fig. 2.
where (kx , k y ) are the two amplification factors of the To make it more clearly, S1 and S2 are placed into the
camera from the image plane to the image coordinates, same image. In the seam image, the intersection point
(u0 , v0 ) is its principal point. of the middle line in the v-axis and the seam line is
The coordinates of P2 in the frame C are trans- defined as the image feature. At the initialization stage,
formed to the coordinates in the frame T by the torch is manually aligned with the initial point P1 .
The corresponding image feature is f2 . At the welding
⎡ ⎤ ⎡ ⎤⎡ ⎤ stage, the position of the seam is changed from S1 to S2 .
xt2 1 00 0 x c2
⎢ yt2 ⎥ ⎢0 10 L⎥ ⎢ y c2 ⎥ The aim of initial point positioning is to automatically
⎢ ⎥=⎢ ⎥⎢ ⎥ (2)
⎣ zt2 ⎦ ⎣0 01 −h⎦ ⎣ zc2 ⎦ adjust the torch to align with P3 . Suppose the target
1 0 00 1 1 image feature of S2 is f4 when the torch is aligned with
P3 , as long as f4 is computed, we can design the control
where L and h are the distance from the origin of system for the initial weld point positioning based on
the frame T to that of the frame C in the Yt and Z t the difference between the target image feature f4 and
directions, respectively. the current image feature of S2 . Thus, the problem
Suppose the coordinates of P1 in the frame T are
(xt1 , yt1 , zt1 ), from the relationship between P1 and P2
in the frame T, we can get
⎡ ⎤ ⎡ ⎤⎡ ⎤
xt2 1 0 0 L tan θ1 xt1
⎢ yt2 ⎥ ⎢0 10 L ⎥ ⎢ yt1 ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ (3)
⎣ zt2 ⎦ ⎣0 01 1 ⎦ ⎣ zt1 ⎦ P1 P2 S2
1 0 00 1 1 P4 S1
P3
where θ1 is the angle between S1 and the Yt -axis.
From Eq. 1 to Eq. 3, the relationship between the (a)
position of the torch and the image feature of the seam
can be computed when the torch is aligned with the
initial weld point, i.e., the relationship between (u2 , v2 )
and (xt1 , yt1 , zt1 ), which is computed as follows:
f4 f2
⎧
⎪ (u2 − u0 )zc2
⎪
⎪ xt1 = − L tan θ1
⎪
⎪ kx
⎨ S2 S1
(v2 − v0 )zc2 (4)
⎪ yt1 =
⎪
⎪ ky
⎪
⎪
⎩
zt1 = zc2 − h (b)
Fig. 2 The relationship between the position of the torch and the
At the welding stage, suppose the position of the seam image. a the two positions of the torch; b the corresponding
seam changes from S1 to S2 , the relationship between seam image
1538 Int J Adv Manuf Technol (2013) 66:1535–1543
of initial weld point positioning is converted to the reference to deduce the geometric relationship. At the
computation of the target image feature of the seam. welding stage, the target image feature is computed
From Fig. 2, it can be seen that there exists geometric based on the geometric relationship between the two
relationship between the two seams S1 and S2 , based seams at the two stages. Then, the control system is
on which we can compute the target image feature designed to make the torch automatically aligned with
f4 . Obviously, when the torch is aligned with P1 and the initial weld point. In the following sections, the two
P3 , the coordinates of the two points in the frame T parts of the initial point positioning are given in detail.
are the same, i.e., xt1 = xt3 , yt1 = yt3 , zt1 = zt3 . The v-
coordinates of image features f2 and f4 are also the
same, i.e., v2 = v4 . From Eqs. 4 and 5, the target image 4.1 The initialization stage
feature f4 can be computed based on the geometric
relationship of S1 and S2 as follows: The initialization stage is conducted only once for weld-
ing the first plate, which is composed of four steps:
u4 = u2 − L ∗ kz ∗ (tan θ1 − tan θ2 ) (1) aligning the torch manually; (2) image feature
(6)
v4 = v2 extraction; (3) sample point acquisition; and (4) com-
putation of the parameters of the seam line. The main
where kz = kx /zc2 . Since planar weld plate is consid- steps of this stage are given as follows.
ered here, the height of the camera from the weld plate
is changeless, i.e., zc2 is a constant. Thus, kz is a constant
in the whole welding process. 4.1.1 Sample point acquisition
Fig. 3 The flow chart of the proposed initial point positioning p(i) = arg min c(k, i), (i = 1, 2, · · · , h) (7)
method k
Int J Adv Manuf Technol (2013) 66:1535–1543 1539
α2
Gray α1
value hw
Cp
the precision of the parameters of the seam line. These 4.2.1 Computation of the target image feature
points need to be eliminated at first, and then the
parameters of the seam line are computed using the The target image feature of S2 is computed using the
remaining valid points. The set of the valid points is geometric relationship of Eq. 6. After the initialization
determined by stage is completed, the parameters u2 and tan θ1 in Eq. 6
have been determined. They are kept the same in the
Sv = X Pi , Y F i | di < αd d (13) whole welding task since the relationship between the
torch and the camera is changeless. The parameter
where Sv is the set of the valid points, di is the distance tan θ2 has been determined after the first three steps of
from the point (X Pi , Y Fi ) to the fitted line, d is the the welding stage. The parameter L is the distance be-
average distance,and αd is a coefficient to compute the tween the torch and the camera, which can be measured
threshold value. di is computed as follows: using the position value of the torch and the camera
read from the rotary encoder. The parameter kz can be
|Y Fi − ks X Pi − b s | simply calibrated using the following method.
di = (14) Put a ruler under the camera, and the image is
1 + k2s
acquired. Select two points on the ruler, the distance
Based on the set of the valid points, the slope of the of the two points in the image can be measured de-
seam line S1 at the initialization stage is computed as noted as Di . The distance of the two points in the
follows: Cartesian space can be read from the ruler directly,
which is denoted as Dc . The parameter kz is computed
Nv as kz = Di /Dc .
(FVi − FV )(PVi − PV )
i=1
tan θ1 = (15)
Nv
(PVi − PV )2 4.2.2 Closed-loop control of the torch alignment
i=1
The diagram of the closed-loop control system of the
where (PVi , FVi ) is the i-th point in the set Sv , Nv is
initial point positioning is shown in Fig. 6, which con-
the number of the valid points, and PV and FV are
sists of three parts such as the computation of the
the average coordinates of the valid points, which are target and the current image features, the forward con-
computed similarly as Eq. 12.
troller part. The target image feature f4 is used as the
reference signal in the control system, and the current
4.2 Welding stage image feature fc as the feedback signal. Both of them
are computed in the computer using the seam images
Welding stage is carried out every time for welding captured from the camera. The time for computing the
every weld plate, which is composed of five steps: target or the current image feature is determined by
(1) image feature extraction; (2) sample point ac- a soft switch, which is realized in the programmable
quisition; (3) computation of the parameters of the logic controller (PLC). When the switch is put on the
seam line; (4) computation of the target image feature; point pr , the target image feature is computed based
(5) and design of the alignment control system. The on the seam image, the position of the torch read from
former three steps are completed similarly as those at the encoder and the geometric model of Eq. 6. On
the initialization stage. Thus, they are omitted here. the other hand, when the switch is put on the point
The latter two steps are given as follows. pc , the current image feature is computed based on
Rotary encoder
Int J Adv Manuf Technol (2013) 66:1535–1543 1541
the seam image. According to the working principle of The parameters used in the image processing and the
the stepper motor, incremental proportional–integral– geometric model were as follows: the size of the image
differential (PID) controller is used to align the torch was 640 × 512 in pixel, α1 = 6, α2 = 8, the time for
based on the difference between the target and the processing one image was about 52 ms, L = 150 mm,
current image features as shown in Eq. 16. The aligning
process adopts the “look-and-move” control model.
The current image feature is computed and compared 63
with the target image feature. If the difference between Sample points
62.5 Fitted line
them is larger than a threshold value, the adjusting con-
tinues. On the other hand, if the difference is smaller 62
60
+ Kd ∗ e(k)−2 ∗ e(k−1)+e(k−2) (16)
59.5
kz = 5.65, N = 14, and αd = 0.8. The parameters in the are shown in Fig. 7. From Fig. 7, it can be seen that the
control system design were set as follows: the sample extracted seam line in the Cartesian space fits the sam-
time was set 60 ms, s = 24 pulse/pixel, the proportional, ple points very well. The parameters of the seam lines
integral, and differential gains of the PID controller at the two stages were computed as −0.0136 and 0.0147,
were set as 0.4, 0.1, and 0.05, respectively. respectively. Using the geometric relationship between
To test the effectiveness of the new method, the the two seams in Eq. 6, the target image feature of the
initialization stage was firstly completed. Then, the current seam line was computed as 376. It can be seen
seam plate was placed with different inclination and the from Table 1 that the current image feature was 392.
initial point positioning function was activated. In one The aim of the initial weld point positioning is to adjust
of the experiments, the 14 sample points collected at the position of the torch to make the corresponding
the initialization stage and the welding stage are shown image feature change from the current image feature
in Table 1. It can be seen that at the initialization stage, to the target one. The results of the torch alignment
the image feature is 352 when the torch is aligned with using the proposed control system are shown in Fig. 8.
the initial weld point. The results of the computation of Figure 8a shows the position error changes of the torch
the parameters of the seam line by line fitting method in the alignment process. Figure 8b shows the output
pulses sent to the stepper motor. It can be seen that
the alignment error of the torch is quickly removed
3
in about 34 control periods, i.e., 2.04 s, which shows
the effectiveness of the proposed alignment control
2.5 system.
Position error of the torch (mm)
1.5
6 Conclusion
1
Acknowledgements This work was supported by the National tion. In: Proceedings of the IEEE international symposium
High Technology Research and Development Program of China on industrial electronics, vol 2, pp 169–185
under grant 2006AA04Z213 and in part by the National Natural 8. Lee SK, Na SJ (2002) A study on automatic seam tracking in
Science Foundation of China under grant 60725309. pulsed laser edge welding by using a vision sensor without an
auxiliary light source. J Manuf Syst 21(4):302–315
9. Fang ZJ, Xu D, Tan M (2010) Visual seam tracking system
for butt weld of thin plate. Int J Adv Manuf Technol 49:519–
526
References
10. Zhang YM, Kovacevic R, Li L (1996) Adaptive control of full
penetration gas tungsten arc welding. IEEE Trans Control
1. Xu YL, Yu HW, Zhong JY, Lin T, Chen SB (2012) Real- Syst Technol 4(4):394–403
time seam tracking control technology during welding ro- 11. Bae KY, Lee TH, Ahn KC (2002) An optical sensing system
bot GTAW process based on passive vision sensor. J Mater for seam tracking and weld pool control in gas metal arc
Process Technol 212(8):1654–1662 welding of steel pipe. J Mater Process Technol 120(1):458–
2. Wu YQ, Yuan ZH, Wang JH (2012) A fuzzy controller design 465
of seam tracking for welding robot. Adv Mater Res 442:370– 12. Zhang YM, Liu YC (2007) Control of dynamic keyhole weld-
374 ing process. Automatica 43(5):876–884
3. Mikael F (2003) Development of intelligent robot systems 13. Xu L, Cao MY, Wang HX, Collier M (2008) A method to
based on sensor control. PhD thesis, Division of Robotics, locate initial welding position of container reinforcing plates
Department of Mechanical Engineering, Lund University using structured-light. In: Proceedings of the 27th Chinese
4. Xu PQ, Xu GX, Tang XH, Yao S (2008) A visual seam track- control conference, Kunming, Yunan, China, pp 310–314
ing system for robotic arc welding. Int J Adv Manuf Technol 14. Zhu ZY, Lin T, Piao YJ, Chen SB (2005) Recognition of the
37:70–75 initial position of weld based on the image pattern match
5. Fang ZJ, Xu D, Tan M (2011) A vision-based self-tuning technology for welding robot. Int J Adv Manuf Technol
fuzzy controller for fillet weld seam tracking. IEEE Trans 26:784–788
Mechatron 16(3):540–550 15. Chen XZ, Chen SB (2010) The autonomous detection and
6. Kuo HC, Wu LJ (2002) An image tracking system for guiding of start weldidng position for arc welding robot. Ind
welded seams using fuzzy logic. J Mater Process Technol Rob 37(1):70–78
120:169–185 16. Chen XZ, Chen SB, Lin T, Lei YC (2006) Practical method
7. Pritschow G, Mueller G, Horber H (2002) Fast and robust im- to locate the initial weld position using visual technology. Int
age processing for laser stripe-sensors in arc welding automa- J Adv Manuf Technol 30:663–668