JOURNAL OF LASER APPLICATIONS

VOLUME 21, NUMBER 4

NOVEMBER 2009

Seam-tracking for high precision laser welding applicationsMethods,

2 restrictions and enhanced concepts
1

Boris Regaard,1,a Stefan Kaierle,2,b and Reinhart Poprawe2

3
4
5

Fraunhofer Center for Laser Technology, Plymouth, Michigan 48170

Fraunhofer Institute for Laser Technology, D-52074 Aachen, Germany

Received 17 October 2008; accepted for publication 20 August 2009

Laser beam welding bears evident advantages regarding precision, quality, productivity, low heat
input, and feasibility of automation. At the same time the process calls for high precision of the
beam positioning on the workpiece, which therefore imposes high requirements of welding
trajectory and feed rate accuracy; e.g., for butt welding the focal point of the laser beam with respect
to the joint must be maintained within a typical accuracy better than 20 150 m, depending on the
focused beam radius. To meet these requirements, seam-tracking devices are used. A sensor
measures the joint position and computes a correction vector to compensate the joint trajectory
offset. The deviation is compensated either by a robot trajectory adjustment or by an additional
tracking axis. This paper describes the basic concepts of seam tracking in detail and points out
problems in the different control principles, which are evoked by the forerun of the sensor.
State-of-the-art sensors and error compensating techniques are presented and analyzed. Further, a
new approach for seam tracking is introduced. It uses a multisensor concept, which in addition to the
seam position measures the relative displacement between the processing head and the workpiece.
An integrated two-dimensional beam positioning system enables self-guided processing, which
allows high-accuracy tracking of a joint independent of the motion system and disengages from time
intensive sensor calibration and robot teaching necessity. 2009 Laser Institute of America.

8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

23 I. INTRODUCTION

Great advantages of laser welding in comparison to arc

25 or resistance welding are the high achievable accuracy and
26 aspect ratio of the weld seam at simultaneously low heat
27 input into the workpiece. However, these advantages at the
28 same time comprise challenges because the thin laser beam
29 needs to be guided on the joint within tight boundaries as
30 small as 20 m typically 50 m in butt welding, fillet
1
31 welding or double flanged seam welding applications. In
32 overlap welding applications, the lateral position accuracy
33 requirement is lower, usually within 0.2 1 mm.
34
The path accuracy of the laser beam with respect to the
35 joint mainly depends on three variables: the robot path accu36 racy, the workpiece geometry, and the repeatability of the
37 workpiece fixture.
38
Standard articulated robots usually achieve good posi39 tion accuracy; however, the path accuracy isdependent on
40 the controller algorithmwithin several millimeters. Lange
2
41 et al. measured a maximum deviation error of 5.37 mm and
42 a rms error of 0.806 mm of a circular path with a standard
3
43 KUKA KR6/1 robot, an rms error of 0.5 to 1.2 mm at dif44 ferent articulated robots. In comparison, a gantry system de45 signed for laser welding applications gains maximum devia4
46 tion errors within 0.15 mm. High efforts are taken to
47 improve the path accuracy of articulated robots using ad48 vanced feed forward control systems, FEA improved control
24

Electronic mail: bregaard@clt.fraunhofer.com

Electronic mail: skaierle@ilt.fraunhofer.de

1042-346X/2009/214/1/0/$25.00

algorithms, and mechanical improvements;2,57 however, articulated robots are not applicable for most butt- or filletwelding applications.
Comparable challenging to the robot path accuracy is the
compliance of the joint position accuracy, imposing high demands on workpiece accuracy and fixture repeatability, in
particular considering workpiece distortion through heat input of the welding process. This is often the reason for process irregularities and weld defects.8
In order to reduce the accuracy requirements of the
workpiece geometry and the fixture and therefore to enable
laser butt welding for mass production industry, seamtracking devices are used. As explained in Sec. III C, standard seam-tracking devices only compensate for joint trajectory of fixture deviations andafter calibrationrepetitious
path deviations of the robot. Deviations in robot repeatability
or thermal movement of the workpiece or fixture can only be
corrected by seam-tracking systems without sensor forerun9
or with a relative movement feedback as described in Sec.
IV D that closes the control loop around the end effector
position.

49

II. SEAM-TRACKING PRINCIPLE

70

The first seam-tracking devices where introduced in the

early 1980s.10,11 They were primarily used in arc welding
applications, which have fewer requirements regarding precision and speed. Generally, a seam-tracking system consists
of a joint position measurement device sensor, a tracking

71
72
73
74
75

2009 Laser Institute of America

50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

FIG. 3. Seam-tracking principle using gray-scale image processing.

FIG. 1. Seam-tracking principle based on a scanning triangulation sensor.

axis linear or rotary, and a control unit. The tracking axis

77 may be abandoned if the robot has on-line path correction
78 capability.
76

79 A. Sensor concept

The predominant seam-tracking sensor concept is based

on the triangulation principle. Older sensors of this type use
82 a deflecting mirror to scan the workpiece surface around the
12
83 joint. The joint position is recognized as a discontinuity in
84 the measured distance between sensor and workpiece surface
85 Fig. 1. Advantages of this concept are as follows:
80
81

approved principle using point-shaped laser beams and line

cameras;
88 fast and simple triangulation algorithm line scan camera,
89 one-dimensional search;
90 adjustable measurement resolution in scanning direction
91 by varying the scanning speed and the sensor reading fre92 quency; and
93 high illumination power of the triangulation laser point94 shaped laser beam.
86
87

The disadvantages are mean robustness due to moveable

parts and a limited temporal resolution.
Nowadays, the latter concept is very rare. Most often
97
98 used are light section sensors, which also utilize the triangu99 lation principle, but stretch it to a second dimension. Instead
100 of a point-shaped triangulation laser beam, a laser line is
101 projected onto the workpiece surface Fig. 2. The detector
102 has two-dimensional resolution CCD or CMOS camera.
95
96

FIG. 2. Seam-tracking principle using a line section sensor.

Advantages of this concept are as follows:

103

robust setup; no movable parts;

104
high temporal resolution possible dependent on the cam- 105
era framerate and image processing algorithm; and
106
reliable joint detection.
107
The measurement resolution in feed direction can be increased by using multiple parallel laser lines. The camera
monitors the parallel lines in one image, which enables to
measure multiple joint positions at different distances simultaneously, which is equitable to an increase of the camera
framerate.13
Seam-tracking sensors using the light section principle
are available, e.g., from Falldorf-Sensor GmbH14 ServoRobot inc.,15 and Precitec KG.16 Multiple lines are utilized, e.g.,
by Meta-Scout GmbH.17
A less used optical measurement concept is gray-scale
image processing.10,18 This sensor type also uses a twodimensional detector camera to observe the workpiece surface. Instead by a well-defined laser line, the joint and workpiece surface is homogeneously illuminated with diffuse
light. The joint position is recognized not by discontinuity of
the laser line but by separating areas of different reflectivity
or brightness within the camera image Fig. 3.
Advantages of this concept are as follows:

108

small sensor design no triangulation angle needed;

simultaneous measurements in different distances possible;
higher measurement reliability;
detection of thin butt joints technical zero gap possible;
and
sensor adjustment in relation to joint direction not relevant.

127
128
129
130
131
132

A disadvantage of this sensor principle is the limited

illumination and observation angle; the inclination to the
workpiece surface normal should not exceed 37, dependent on the surface finish and material.
Seam-tracking sensors using this principle are available
e.g. Plasmo Position controller19. Trumpf Lasertechnik
GmbH is also using gray-scale image processing in seamtracking applications.
Besides these optical seam-tracking sensor principles,
mechanical sensors can be found in industrial
applications.11,20,21 These sensors consist of a well-defined
tip or wheel, or they utilize existent components, e.g., the
filling wire or the pressure wheel,22 for seam-tracking Fig.

133
134
135
136

109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126

137
138
139
140
141
142
143
144
145

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

direction of travel
joint
sensor

TCP
welding
head

corr

tracking
axis

FIG. 4. Mechanical seam-tracking principle using a tip, the filler wire, or a

tracking wheel.

4. Pressure or elongation sensors at the mounting indicate

147 deviations of the joint and give feedback to the controller
148 active seam tracking. Rare applications use the mechanical
149 guidance force of the joint edge onto a tracking wheel di20
150 rectly to follow the joint. In this case, no controller is used
151 passive seam-tracking.
152
Advantages of mechanical sensors are as follows:
robust cheap setup;
154 small integrated sensor; existing components may be alien155 ated;
156 good dirt resistance, no optical parts.
153

Mechanical sensors are only usable for contoured joints

like fillet welds or Y-shaped butt welds; I-shaped butt welds
are not detectable. Furthermore, the joint has to be fairly
17
160 linear. A disadvantage is possible tip abrasion, which leads
161 to measurement errors. The tip also can damage the work162 piece surface and abets collision risks.
157
158
159

robot
hand
TCP

x
y

146

joint

robot

robot arm

FIG. 6. Seam-tracking setup with sensor fixed to the TCP closed loop
configuration.

robot hand. The welding head with the TCP tool center
point; the point of incident of the welding laser sits on a
tracking axis, which allows adjusting the TCP relative to the
robot hand Fig. 5. The tracking axis is usually a precise
and fast linear axis, but can also be a rotary axis.23
In a more common setup the sensor is fixed to the
welding head Figs. 6 and 7. The advantage of this setup is
that the sensor moves with the TCP. Since the TCP matches
the current joint position, the sensor is continuously
readjusted and therefore covers a greater deviation between
joint and robot trajectory.24 The tracking axis may be

163 B. Sensor arrangement

164
165

Two possible arrangements are used for seam tracking.

In a less common setup, the sensor is installed fixed to the

FIG. 5. Seam-tracking setup with sensor fixed to the robot hand open loop
configuration.

FIG. 7. Principle of seam-tracking systems.

166
167
168
169
170
171
172
173
174
175
176

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

rtarget,yt = rjoint,yrjoint,x=rTCP,xt = syt T + rTCP,yt T.

3 213
The delay depends on the forerun of the sensor sx and the 214
average feed rate used to bridge the forerun distance. In the 215
216
case of a constant feed rate rrobot,x, the delay equals
T = sx/rrobot,xrrobot,x=const .

4 217

The current TCP position in the global coordinate system r 218

219
depends on the robot hand and tracking axis position,
rTCP,yt = rrobot,yt + paxis,yt.

5 220

The nominal position of the tracking axis pcorr,y equals the 221
difference of the target position rtarget,y and the robot 222
position rrobot,y. With formulas 3 and 5 it can be 223
224
calculated to

FIG. 8. Vector definition, closed loop setup.

pcorr,yt = rtarget,yt rrobot,yt

177

abandoned and replaced by direct robot trajectory

178 adjustment; this corresponds to the latter setup type. To
179 reduce the sensor forerun and to reduce size, the sensor can
180 be integrated into the welding head observing the workpiece
23,25
181 coaxial to the laser beam.
From the control point of view, the first setup type is an
182
183 open loop control, while the latter setup is closed loop. Due
184 to its obvious advantage of a larger acceptance width; the
185 latter principle is used more often; however, open loop
186 control is the more robust principle and provides easier
187 setup.

188 C. Control principle

Due to its more common use, the control principle

190 described in the following refers to the closed loop control
191 configuration. The difference to the open loop configuration
192 Fig. 5 is that the joint position measured by the sensor s y is
193 relative to the actual TCP position rTCP
189

194
195
196
197

s = rjoint rTCP ,

whereas in open loop configuration it is relative to the robot

hand position rrobot;
s = rjoint rrobot .

Since the TCP position, in comparison to the robot position,

is controlled by the sensor measurand, this configuration is
closed loop.
Figure 8 illustrates the vector system used in the
201
202 following. The coordinate system r is fixed to the workpiece;
203 the coordinate system p is fixed to the robot hand. paxis,y
204 determines the actual tracking axis position. The sensor
205 coordinate system s has its origin in the TCP; sx specifies the
206 sensor forerun, s y the measured joint position relative to the
207 TCP.
As a matter of principle, the sensor measures the joint
208
209 position not in the TCP position, but with a forerun of
210 typically 40 200 mm. Therefore, a time delay has to be
211 considered between the measuring of a joint trajectory
212 deviation and its correction,
198
199
200

225

= syt T + rrobot,yt T + paxis,yt T

rrobot,yt.

226

6 227

Usually, the lateral movement of the robot hand rrobot,y is not

measurable and assumed to be zero within the accuracy
requirements. Without lateral robot movement the
nominal axis position can be calculated by the sensor
measurand, the axis position and the time delay,
pcorr,yt = syt T + paxis,yt Trrobot,y=const .

228
229
230
231
232

7 233

Figure 9 shows the simplified action diagram of the control

principle according to formula 7.
This control principle requires that the current tracking
axis position paxis,y is measurable. If this is not possible, the
nominal tracking axis position pcorr,y may be used
alternatively; however, the contouring error induces
additional deviations. They can be reduced by using a
dynamic model of the tracking axis.26 This substitution can
be used in all control principles described in the following.
Simpler seam-tracking systems ignore the forerun and
use a conventional PID control principle to control the
correction axis,27 which leads to principal positioning errors.
Pritschow et al.28 quantify these errors in relation to TCP
velocity, sensor forerun, control parameters, and the joint
trajectory.
A seam-tracking concept from Thyssen Krupp AG25 also
ignores the forerun but claims that the residual error is not
relevant for the welding application due to a small sensor
distance to the TCP of about 2 mm. This is true if the actual
delay of the controller matches the necessary delay
according to formula 4.

234
235

III. ERROR SOURCES IN SEAM-TRACKING

APPLICATIONS

255
256

236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254

The seam-tracking control principle shown in the previ- 257

ous chapter implicates the assumption of a constant linear 258
robot movement, requiring
259
a constant feed rate;
no lateral movement of the robot hand;

260
261

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

FIG. 9. Simplified seam-tracking control principle for closed loop configuration, not considering rotation and lateral robot movement.

However, in real applications lateral displacement does 290

291
occur, for example caused by

262

no rotation of the welding head or robot hand; and

263 no movement or distortion of the workpiece.
264
265
266

In numerous real applications, these requirements are not

given, thus leading to positioning errors, which will be analyzed in the following.

acceleration limitations of articulated arm robots;

deliberate lateral movement or rotation in 2D applications;
vibration; and
thermal distortion.

292
293
294
295

A hereby caused deviation is measured by the sensor but not 296

267 A. Error type 1: Feed rate variation

As shown in formula 4, the time delay T between the

measuring of a joint deviation and its correction depends on
270 the sensor forerun and the feed rate of the robot system. The
271 sensor forerun is given by the mechanical setup, but the feed
272 rate eventually differs, provoked by
268
269

276
273
274
275

277
278
279
280

281
282
283
284

deceleration on a small curve radius or edge;

axis shift or inaccuracy of the robot;
thermal distortion; and
processing conditions.

The result is an abridged TCP trajectory that leadsin the

case of a nonlinear joint trajectoryto a positioning mismatch Fig. 10. Its length in feed direction can be determined by

feedrate,x = rtarget,x rTCP,x = sx 1

rrobot,x,real
,
rrobot,x,exp

where rrobot,x,exp is the expected and rrobot,x,real is the real

feed rate. The absolute mismatch depends on the actual
joint contour.

285 B. Error type 2: Lateral robot movement

286
As shown in Fig. 9, the lateral movement of the robot
287 rrobot,y is assumed to be zero in the control algorithm not
288 considering error correction techniques described later in
289 this article.

FIG. 10. Positioning error caused by varied feed rate.

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

FIG. 11. Positioning error caused by lateral displacement of the robot

system.
FIG. 13. Positioning error caused by rotation of the welding head.

297

corrected instantly: due to the sensor forerun sx a time delay

is considered, although instant correction would be required,
299 cf. formula 6.
The result is a temporary position mismatch, comparable
300
301 with a control delay. The length of the positioning mismatch
302 equals the sensor forerun Fig. 11; its value equals the lat303 eral movement of the robot,
298

displacement,y = rtarget,y rTCP,y = rrobot,yt T rrobot,yt.

9 304

C. Error type 3: Rotation of the welding head

305

Seam-tracking systems are extremely sensitive to

rotation of the welding head relative to the feed direction.
The rotation provokes a lateral deviation of the TCP
trajectory and at the same time a shortening of the sensor
forerun related to the feed direction,

306
307
308
309
310

rTCPt = rrobott + R p axist,

10 311

st = R1 rjointrjoint,x=rTCPt + R stx rTCP,yt,

11 312

with

R=

FIG. 12. Deviation of the TCP position rotation caused by welding head
rotation .

313

cos sin
sin

cos

12

314

Both modifications are not measured by the sensor and 315

cause a constant position mismatch of the TCP Fig. 12.
316

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

TABLE I. Error correction techniques.

A: Dynamic
time delay
Requirements
Feed rate
measurable

High robot time

accuracy

320
321
322

4: Workpiece
movement

The lateral positioning error at a linear gap Fig. 13

corresponds to
rotation,y = rtarget,y rTCP,y = sx sin,

13

where is the welding head rotation. The shortening in feed

direction can be determined by
rotation,x = rtarget,x rTCP,x = sxcos 1.

14

A movement of the workpiece within the process time

span is usually caused by thermal distortion of the
workpiece or the fixture through local or global heat input.
The movement can be parallel or lateral to the robot
movement or can induce rotation relative to the welding
head, thus causing the same results as error types 13. In
difference to path deviations induced by the robot, a
workpiece movement cannot be corrected by conventional
error correction techniques, which read the robot positioning
data and have no information about the workpiece position.

types is to abridge the sensor offset sx. A significant shortening is realized with a coaxial sensor setup, e.g., Refs. 23, 25,
and 29.
For fillet or flange welds, Jackel et al.21 use the filler
wire instead of an optical sensor for seam tracking. The filler
wire receives a force if abutted on the joint edge, which is
gauged to control the correction axis. Since the tip of the
filler wire is coincident with the laser TCP, this principle
possesses zero sensor forerun, avoiding all previous described seam-tracking errors.
If a sensor forerun cannot be avoided, production capable seam-tracking sensors use different correction tech-

tracking
axis

paxis , y

(no correction)

rTCP , y (rTCP , x )

s y (rTCP , x + s x )

r jo int, y (rTCP , x + s x )

rrobot , y

334 IV. ERROR CORRECTION TECHNIQUES

All errors described in the previous section are originated in the sensor forerun with respect to the TCP. Therefore, a way to reduce sensor-originated deviation in all error

joint
trajectory

robot

scal,y

335
336
337

3: Rotation of
welding head

323 D. Error type 4: Workpiece movement

324
325
326
327
328
329
330
331
332
333

2: Lateral robot
movement

319

Reference workpiece
with ideal contour

318

D: Self-guided
processing

Corrected error
1: Variation of feed
rate

C: Two-sensor position
measurements

High robot repeat

accuracy

Robot-sensor
calibration/sync

317

B: Calibration on
reference workpiece

(ideal contour)

tracking
sensor

Calibration vector
(stored for work run)

FIG. 14. Storing of the calibration vector Scalt using a reference workpiece
with ideal joint trajectory rjoint,y = 0.

338
339
340
341
342
343
344
345
346
347
348
349

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

A. Dynamic time delay

362

If the actual feed rate is measurablee.g., provided by

the robot controlit can be considered in the control
algorithm by implementing a variable delay T to correct
error type 1. The variable time delay T has to meet the
restriction

363

350

niques to enable seam tracking for real world applications.

351 Table I lists the most common correction techniques, their
352 restrictions, and the error types they compensate.
353
Usually, method A dynamic time delay is combined
354 with method B calibration on reference workpiece or
355 method C double joint position measuring. If single error
356 types can be precluded by mechanical assumptions or setup
357 arrangements, individual error correction techniques may be
358 obsolete.
359
Method D self-guided processing is capable to correct
360 all four error types without needing a high robot accuracy,
361 calibration or robot-sensor synchronization.

365
366
367

t=tT

FIG. 15. Correction of error type 2 using a calibration vector Scalt.

364

rrobot,xt = sx .

15
368

B. Calibration on reference workpiece

369

A commonly used way to minimize error type 2 is to

gather a calibration vector Scalt along the welding path.24
Usually robots have a poor absolute accuracy but a
comparatively good repeat accuracy. In a dry run, the sensor
measures the deviation to a reference workpiece, which
features a joint deviation to the target path less or equal the
needed accuracy. If the sensor measures a deviation, this
deviation is originated in a lateral movement of the robot
Fig. 14.
In a work run the calibration value is time-equidistant
subtracted from the measurand to correct the robot path
deviation Fig. 15,

370
371

rcorr,yt = syt T + rcorr,yt T scal,yt.

373
374
375
376
377
378
379
380
381

16 382

The time-dependent sensor calibration eliminates errors

caused by lateral robot movement if an exact timing between
the robot movement and the sensor calibration and a good
robot repeat accuracy can be ensured. However, the costs to
produce a reference workpiece and the effort of calibration
have to be considered.

FIG. 16. Correction of error type 2 using a second sensor s2 with a different forerun distance s2,x s1,x.

372

383
384
385
386
387
388

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

389 C. Two-sensor position measurement

p corrt = rtargett + p axist rTCPt.

For correction of error type 2 without calibration on-line

measuring of the lateral robot movement is required.
30
392 Falldorf measures the lateral movement by calculating the
393 difference of the joint-to-TCP deviation at two unequal
394 forerun distances s1x and s2,x s2,x s1,x at the same time.
395 The difference of the measurands at equal TCP positions
396 rTCP,x is originated in a lateral TCP movement rTCP,y ,
390
391

397
398
399

rTCP,yt = rTCP,yt Ts + s2,yt Ts s1,yt,

17

with
Ts = s2,x s1,x/rrobot,x .

Knowing both the TCP and the joint trajectory, the lateral
401 TCP position rTCP,y can be set onto the corresponding
402 lateral joint position Fig. 16. This correction technique
403 requires a known or constant feed rate and no welding
404 head rotation.

405 D. Self-guided processing

The four error types described in Sec. IV can be all

attributed to a lack of information. The seam-tracking sensor
408 only measures the joint position on the workpiece relative to
409 the current TCP position; however, the actual position of the
410 TCP relative to the workpiece is not available. To calculate
411 the TCP trajectory with this incomplete information,
412 irregularities besides deviation of the joint, particularly
413 deviation of the robot trajectory, welding head orientation
414 and displacement of the workpiece, have to be assumed to
415 be not existent. The correction of the sensor offset has to be
416 carried out time-based instead of position based; cf. formula
417 6.
418
The lack of information is compensated with methods B
419 and C by measuring an error value either by means of a
420 separate calibration run or a second joint position
421 measurement. Both methods can increase tracking accuracy;
422 however, additional assumptions have to be assured cf.
423 Table I.
424
For self-guided processing, the idea is to actually
425 measure the displacement between the TCP and the
426 workpiece. This additional information enables to calculate
427 both the TCP trajectory and the joint trajectory in a
428 coordinate system stationary to the workpiece,
406
407

rTCPt = rconst +

With this concept, seam tracking is performed twodimensional position based instead of time based. All four
error types described in Sec. IV are inexistent.
The concept can be enhanced by a feedback of the target
position to the robot control dotted line in Fig. 17. With
this feedback, seam tracking of unknown joint trajectories is
feasible. In this case, the high accuracy positioning is
performed by the two dimensional tracking axis, while the
rough contour is followed by the robot.

437

V. REALIZATION OF A SELF-GUIDED PROCESSING

HEAD

446
447

A. Displacement sensor

448

A sensor principle to measure the relative displacement

to a workpiece surface is described in Ref. 34. It utilizes the
rough surface finish of the workpiece: the surface structure
is unique and stationary on the workpiece. A relative
displacement between the sensor and the workpiece
therefore can be measured by finding the maximum cross
correlation of areas in consecutive observed images Fig.
18, a principle also known as optical flow.32 The relative
velocity corresponds to the displacement normalized by the
frame rate of the camera.
The developed sensor consists of a high speed camera
CMOS camera, which observes the workpiece vertical,
preferable coaxial to the laser beam to realize a short forerun
and secure installation. The workpiece surface is illuminated
coaxially to the camera with a low power illumination laser
Fig. 19.
The camera observes an area of approximately
6 6 mm2. Using narrow band filters and a small and high
intense illumination spot, the thermal radiation of the
welding process and back reflection of the laser beam can be
suppressed completely Fig. 20.

449
450

joint
trajectory

robot
r
rrobot

tracking
axis

r
paxis

+
+

19

r
r joint

r
rTCP

displacement
sensor

r
pcorr

r TCPt dt,

431
432
433
434
435

rjointt = rTCPt + st.

20

Knowing both the TCP and the joint trajectory, the lateral
TCP position rTCP,y can be set onto the corresponding
lateral joint position,
rtargett = rjointirjoint,xi=rTCP,xt .
The correction vector equals the vector difference,

21

r
s joint

rtarget
r

r joint

r
r
rtarget (t ) = r joint (i )

tracking
sensor

r&TCP

430

22 436
438
439
440
441
442
443
444
445

18

400

429

rTCP

r joint , x (i )= rTCP , x (t )

FIG. 17. Correction of all four error types using an additional sensor to track
the relative displacement to the workpiece self-guided processing.

451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469

10

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

FIG. 18. Measurement of relative displacement between consecutive images by finding the maximum cross correlation of areas in consecutive observed
images.

470 B. Joint position measurement

471
472
473
474
475
476
477
478
479

For joint measurement, the same images are utilized as

for the displacement measurement. The technique of joint
recognition by gray-scale image shape analysis is well
known.18 If necessary, light section measurement could be
integrated using an additional illumination. However, the
gray-scale image processing developed for this application
turned out to be stable at different materials as stainless
steel, mild steel, copper, and aluminium alloys with punched
as well as laser cut edges.

480 C. Results

The sensor has been adapted to a 6 kW CO2 scanner

system scan field 40 40 mm2 with a forerun to the TCP
of 200 mm noncoaxial setup. The motion system is an
articulated robot with comparatively low path accuracy.
Tests where performed on mild steel with a sinus-formed
butt joint Fig. 21.
The system is able to follow the joint with randomly
487
488 varying feed rates independent of the welding head
489 orientation and the robot movement. Due to performance
490 limitations of the control system, the maximum feed rate of
491 the test setup is limited to 5 m / min.
481
482
483
484
485
486

D. Further development

492

The self-guided processing optic is currently enhanced

to real two-dimensional seam-tracking by integrating the
sensor as well a 2D positioning system scanner or
motorized adjustable mirror into the optical path of the laser
beam coaxial setup, Fig. 22.
This setup allows to follow a two-dimensional joint
without rotating the welding head. The self-guided optic
only needs to be moved within the scanner field upon the
joint and the laser beam finds its way. Even hand guided
or vehicle hooked up laser welding can be realized for
butt-welding and other high accuracy applications.
Another field of current research is the analysis of
additional information of the coaxial camera sensor for
further process information melt pool geometry, splatter,
and seam geometry, which enables seam tracking and
process monitoring with one single sensor system. The
coaxial illumination of the process zone allows robust
feature detection.33,34

493

VI. CONCLUSION

511

494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510

Seam-tracking sensors are useful tools for laser welding, 512

if the workpiece or fixture accuracy does not meet the needed 513

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

11

FIG. 19. Optical setup of the combined displacement and seam-tracking sensor left: stand-alone sensor and right: coaxial setup.

514

restrictions. However, due to the necessary sensor forerun,

515 conventional control principles based on PID control have
516 limited accuracy.
517
These tracking deviations can be eliminated using an
518 advanced control principle, which takes the forerun into ac519 count. Nowadays, this is state of the art in most commer520 cially available seam-tracking systems. Yet, additional devia521 tions occur through deviation in the feed rate, a lateral
522 movement of the robot hand, rotation of the processing head,
523 or movement or distortion of the workpiece. Different error
524 correction techniques are used to reduce these deviations;
525 however, they all limit flexibility, are time consuming, and
526 need advanced interfacing with the robot system, thus in-

FIG. 20. Observed workpiece surface and welding process coaxial setup.

creasing setup and running costs. Also, standard articulated

robots cannot maintain the required path accuracy and repeatability restrictions, therefore more expensive and less
flexible gantry systems, tricept robots, or optimized articulated robots have to be used.35,36
To eliminate the identified error sources, a seam-tracking
principle and device has been developed, which computes an
error-free correction vector independent of the robot and
workpiece movement. It consists of a multifunctional sensor,
which additionally to the seam position measures the displacement between the processing head and the workpiece.
This information is used by an advanced control principle to
determine the trajectory of the joint as well as the TCP in a
global coordinate system. Therefore, seam tracking can be
performed two-dimensional position-based instead of onedimensional time based.

FIG. 21. Self-guided seam tracking of a sine-wave formed joint.

527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542

12

J. Laser Appl., Vol. 21, No. 4, November 2009

Regaard, Kaierle, and Poprawe

feedrate,x deviation between desired and actual TCP

position caused by feed rate variation in feed
direction
displacement,y lateral deviation between desired and actual
TCP position caused by lateral movement of
the robot hand
rotation,x deviation between desired and actual TCP
position caused by rotation of the welding
head in feed direction
rotation,y lateral deviation between desired and actual
TCP position caused by rotation of the welding head
1

FIG. 22. Coaxial setup of camera sensor, illumination, and laser positioning
system.

543

A concept is introduced, which combines a coaxial measurement concept with a laser positioning system. A processing head using this concept is capable to follow joints or
546 traces on the workpiece self-guided, which means indepen547 dent of the actual movement of the processing head and
548 without any prior calibration or information of the trajectory.
549 The absolute position accuracy is independent from the ac550 curacy of the motion system, which enables applications
551 with low accuracy robots or even hand moved systems.
544
545

552 Nomenclature
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578

paxis,x distance of the TCP to the robot hand center

point in feed direction
paxis,y lateral position of the TCP relative to the
robot hand center point actual tracking axis
position
pcorr,y joint position lateral to TCP relative to the
robot hand center point desired tracking
axis position
rTCP tool center point in global coordinate system
rjoint joint position lateral to sensor position in
global coordinate system
rrobot robot hand center point in global coordinate
system
rrobot,x feed rate of the robot hand
rrobot,y lateral movement of the robot hand
rtarget joint position lateral to TCP in global coordinate system
Sx forerun of the sensor to TCP closed loop or
robot hand open loop
Sy lateral position of the joint relative to the
sensor at sensor position
Scal,y calibration value for lateral sensor position
to correct lateral movement of the robot
hand
T time delay to compensate sensor forerun

M. Schultz, Fertigungsqualitt beim 3D-Laserstrahlschweien von

Blechformbauteilen, Dissertation, University Erlangen-Nuernberg, Germany, 1997, p. 58.
2
F. Lange and G. Hirzinger, Adaptive minimization of the maximal path
deviations of industrial robots, Proceedings of the European Control
Conference, Karlsruhe, VDI/VDE GMA, 1999.
3
F. Garnich, IWB-Forschungsberichte, Laserbearbeitung mit Robotern
Springer, Berlin, 1992, p. 50.
4
P. Hoffmann, Verfahrensfolge Laserstrahlschneiden und-schweien,
Fertigungstechnik Erlangen C. Hanser, Muenchen, 1992.
5
M. Loo, Y. Hamidieh, and V. Milenkovic, Generic path control for robot
applications, Robots 14 Conference Proceedings Society of Manufacturing Engineers, Detroit, 1990, 116.
6
G. Pritschow, M. Bauder, and A. Horn, Erhhung der Bahngenauigkeit
von Industrierobotern, Robotersysteme 8, 162170 1992.
7
A. Albu-Schffer, Regelung von Robotern mit elastischen Gelenken am
Beispiel der DLR-Leichtbauarme, 2002.
8
A. Reek, Strategien zur Fokuspositionierung beim Laserstrahlschweien, Ph.D. thesis, H. Utz Verlag, Mnchen, 2000.
9
L. Weiss, A. Sanderson, and C. Neuman, Dynamic sensor-based control
of robots with visual feedback, IEEE J. Rob. Autom. 3, 404417 1987.
10
G. Starke, Nahtfhrungssensor zur adaptiven Steuerung von
Handhabungseinrichtungen zum Lichtbogenschweien, Dissertation,
University RWTH Aachen, 1983.
11
K. Fuchs, Flexible, sensorgesteuerte Roboterschweisysteme, Aachen,
Fotodruck Mainz, 1987.
12
H. Bgel, Laser fhrt Schweiroboter mit Pilotsensor, Laser-Praxis 1,
6768 1990.
13
W. Trunzer, H. Lindl, and H. Schwarz, Einsatz eines voll 3D-fhigen
schnellen Sensorsystems zur Nahtverfolgung beim Laserstrahlschweien, Laser in der Technik, edited by W. Waidelich Springer, Berlin,
1993, pp. 405410.
14
Falldorf Sensor GmbH, Falldorf Sensor, 2007 http://www.falldorfsensor.de.
15
ServoRobot Inc., Seam finding, Seam Tracking ROBO-TRAC, 2007
http://www.servorobot.com/english/Manufacturing_solutions/robotrac.htm.
16
Precitec KG, Laser Welding-Monitoring-Optical Seam Tracking Using
Light Stripe Sensor, 2007 http://www.precitec.de/precitec/enxml/
products/lpf.html.
17
Meta-Scout GmbH, SCOUT Nahtverfolgung, 2007 http://www.scoutsensor.com/.
18
A. Horn, Optische Sensorik zur Bahnfhrung von Industrierobotern mit
hohen Bahngeschwindigkeiten, ISW Forschung und Praxis Springer,
Berlin, 1994, p. 103.
19
Plasmo Industrietechnik GmbH, Plasmo Industrietechnik GmbH, 2007
http://www.plasmo.eu.
20
I. Haschke, Verfahren und Vorrichtung zum Fgen von Werkstckteilen
mittels eines Energiestrahls, insbesondere Laserstrahls, Patent No.
DE10006852 August 26, 2004.
21
T. Jckel, F. Hanschmann, and S. Schiwy, Process and device for joining
at least two workpieces with laser beam and filter material, the latter used
as contact element for guiding along the joining line, European Patent
No. 06022685 June 27, 2007.
22
HIGHYAG Lasertechnologie GmbH, 2007 www.highyag.de.
23
Permanova Lasersystem AB, 2007 www.permanova.se.
24
J.-P. Boillot, Robot feature tracking devices and methods, U.S. Patent
No. 6,430,472 December 20, 1999.

579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648

J. Laser Appl., Vol. 21, No. 4, November 2009

649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668

25

Thyssenkrupp Steel AG, F. Behr, E. Blumensaat, C. Dornscheidt, M.

Koch, J. Plha, S. Wischmann, L. Ott, andA. Schfer, Patent Application
No. WO/2007/088122 August 9, 2007.
26
R. Kram, U. Kreienkamp, and R. Friedrich, Robotersteuerungsfunktionen und Schnittstellen zur Ankopplung vorlaufender Seneoren, Intelligente Sensorsysteme der Fertigungstechnik, edited by J. Rogos
Springer, Berlin, 1989, pp. 117125.
27
G. Gruhler, Sensorgefhrte Programmierung bahngesteuerter Industrieroboter, SW Forschung und Praxis Springer, Berlin, 1987, p. 67.
28
G. Pritschow and A. Horn, Dynamik derzeitiger Sensorregelkreise fr
Industrieroboter, Robotersysteme 7, 178184 1991.
29
J. Mller-Borhanian, Integration optischer Messmethoden zur Prozesskontrolle beim Laserstrahlschweissen INESS, Abschlussbericht zum
Verbundprojekt Forschungs- und Tagungsberichte, Mnchen, 2005.
30
H. Falldorf, Nahtverfolgung fr das Laserstrahlschweissen unter besonderer Bercksichtigung eines vom Schweissautomaten unabhngigen Systems, Ph.D. thesis, Fortschritt-Berichte VDI. Reihe 8, Mess-,
Steuerungs-und Regelungstechnik, 1995.
31
B. Regaard, S. Kaierle, W. Schulz, and A. Moalem, Advantages of external illumination for monitoring and control of laser materials process-

Regaard, Kaierle, and Poprawe

13

ing, Proceedings of ICALEO 2005, Miami. FL, October 31November

3, 2005, pp. 915919.
32
S. S. Beauchemin and J. L. Barron, The computation of optical flow,
ACM Comput. Surv. 27, 433466 1995.
33
W. F. Clocksin, J. S. E. Bromley, P. G. Davey, C. G. Morgan, and A. R.
Vidler, An Implementation of Model-Based Visual Feedback for Robot
Arc Welding of Thin Sheet Steel, Int. J. Robot. Res. 4, 1326 1985.
34
J. Gedicke, B. Regaard, A. Gillner, and S. Kaierle, Kontrolle beim Mikroschweien: Automatisierte Prozessberwachung durch koaxiale Prozesskontrolle mit Fremdbeleuchtung, Laser-Technik-Journal 3, 3337
2006.
35
B. Regaard, S. Kaierle, and R. Poprawe, Error detection in lap welding
applications using on-line melt pool contour analysis by coaxial process
monitoring with external illumination, Lasers in Manufacturing 2007:
Proceedings of the Fourth International WLT-Conference Lasers in Manufacturing 2007, pp. 471476.
36
A. Huwer, Sensorsystem zur Erfassung variabler Fgespaltweiten beim
Laserstrahlschweien im Stumpfsto, Aachener Berichte Fgetechnik
Shaker, Aachen, 1993.

669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687