Seam-tracking for high precision laser welding applications—Methods,restrictions and enhanced concepts
Boris Regaard,1,a� Stefan Kaierle,2,b� and Reinhart Poprawe2
1Fraunhofer Center for Laser Technology, Plymouth, Michigan 481702Fraunhofer 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 heatinput, and feasibility of automation. At the same time the process calls for high precision of thebeam positioning on the workpiece, which therefore imposes high requirements of weldingtrajectory and feed rate accuracy; e.g., for butt welding the focal point of the laser beam with respectto the joint must be maintained within a typical accuracy better than 20–150 �m, depending on thefocused beam radius. To meet these requirements, seam-tracking devices are used. A sensormeasures the joint position and computes a correction vector to compensate the joint trajectoryoffset. The deviation is compensated either by a robot trajectory adjustment or by an additionaltracking axis. This paper describes the basic concepts of seam tracking in detail and points outproblems 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, anew approach for seam tracking is introduced. It uses a multisensor concept, which in addition to theseam position measures the relative displacement between the processing head and the workpiece.An integrated two-dimensional beam positioning system enables “self-guided” processing, whichallows high-accuracy tracking of a joint independent of the motion system and disengages from timeintensive sensor calibration and robot teaching necessity. © 2009 Laser Institute of America.
I. INTRODUCTION
Great advantages of laser welding in comparison to arcor resistance welding are the high achievable accuracy andaspect ratio of the weld seam at simultaneously low heatinput into the workpiece. However, these advantages at thesame time comprise challenges because the thin laser beamneeds to be guided on the joint within tight boundaries assmall as 20 �m �typically 50 �m� in butt welding, filletwelding or double flanged seam welding applications.1 Inoverlap welding applications, the lateral position accuracyrequirement is lower, usually within 0.2–1 mm.
The path accuracy of the laser beam with respect to thejoint mainly depends on three variables: the robot path accu-racy, the workpiece geometry, and the repeatability of theworkpiece fixture.
Standard articulated robots usually achieve good posi-tion accuracy; however, the path accuracy is—dependent onthe controller algorithm—within several millimeters. Langeet al.2 measured a maximum deviation error of 5.37 mm anda rms error of 0.806 mm of a circular path with a standardKUKA KR6/1 robot,3 an rms error of 0.5 to 1.2 mm at dif-ferent articulated robots. In comparison, a gantry system de-signed for laser welding applications gains maximum devia-tion errors within 0.15 mm.4 High efforts are taken toimprove the path accuracy of articulated robots using ad-vanced feed forward control systems, FEA improved control
algorithms, and mechanical improvements;2,5–7 however, ar-ticulated robots are not applicable for most butt- or fillet-welding applications.
Comparable challenging to the robot path accuracy is thecompliance of the joint position accuracy, imposing high de-mands on workpiece accuracy and fixture repeatability, inparticular considering workpiece distortion through heat in-put of the welding process. This is often the reason for pro-cess irregularities and weld defects.8
In order to reduce the accuracy requirements of theworkpiece geometry and the fixture and therefore to enablelaser butt welding for mass production industry, seam-tracking devices are used. As explained in Sec. III C, stan-dard seam-tracking devices only compensate for joint trajec-tory of fixture deviations and—after calibration—repetitiouspath deviations of the robot. Deviations in robot repeatabilityor thermal movement of the workpiece or fixture can only becorrected 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 effectorposition.
II. SEAM-TRACKING PRINCIPLE
The first seam-tracking devices where introduced in theearly 1980’s.10,11 They were primarily used in arc weldingapplications, which have fewer requirements regarding pre-cision and speed. Generally, a seam-tracking system consistsof a joint position measurement device �sensor�, a tracking
a�Electronic mail: [email protected]�Electronic mail: [email protected]
JOURNAL OF LASER APPLICATIONS VOLUME 21, NUMBER 4 NOVEMBER 2009
1
2
345
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
1042-346X/2009/21�4�/1/0/$25.00 © 2009 Laser Institute of America1
axis �linear or rotary�, and a control unit. The tracking axismay be abandoned if the robot has on-line path correctioncapability.
A. Sensor concept
The predominant seam-tracking sensor concept is basedon the triangulation principle. Older sensors of this type usea deflecting mirror to scan the workpiece surface around thejoint.12 The joint position is recognized as a discontinuity inthe measured distance between sensor and workpiece surface�Fig. 1�. Advantages of this concept are as follows:
• approved principle using point-shaped laser beams and linecameras;
• fast and simple triangulation algorithm �line scan camera,one-dimensional search�;
• adjustable measurement resolution in scanning directionby varying the scanning speed and the sensor reading fre-quency; and
• high illumination power of the triangulation laser �point-shaped laser beam�.
The disadvantages are mean robustness due to moveableparts and a limited temporal resolution.
Nowadays, the latter concept is very rare. Most oftenused are light section sensors, which also utilize the triangu-lation principle, but stretch it to a second dimension. Insteadof a point-shaped triangulation laser beam, a laser line isprojected onto the workpiece surface �Fig. 2�. The detectorhas two-dimensional resolution �CCD or CMOS camera�.
Advantages of this concept are as follows:
• robust setup; no movable parts;• high temporal resolution possible �dependent on the cam-
era framerate and image processing algorithm�; and• reliable joint detection.
The measurement resolution in feed direction can be in-creased by using multiple parallel laser lines. The cameramonitors the parallel lines in one image, which enables tomeasure multiple joint positions at different distances simul-taneously, which is equitable to an increase of the cameraframerate.13
Seam-tracking sensors using the light section principleare available, e.g., from Falldorf-Sensor GmbH14 ServoRo-bot 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-scaleimage processing.10,18 This sensor type also uses a two-dimensional detector �camera� to observe the workpiece sur-face. Instead by a well-defined laser line, the joint and work-piece surface is homogeneously illuminated with diffuselight. The joint position is recognized not by discontinuity ofthe laser line but by separating areas of different reflectivity�or brightness� within the camera image �Fig. 3�.
Advantages of this concept are as follows:
• 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.
A disadvantage of this sensor principle is the limitedillumination and observation angle; the inclination to theworkpiece surface normal should not exceed 3°–7°, depen-dent on the surface finish and material.
Seam-tracking sensors using this principle are available�e.g. Plasmo Position controller19�. Trumpf LasertechnikGmbH is also using gray-scale image processing in seam-tracking applications.
Besides these optical seam-tracking sensor principles,mechanical sensors can be found in industrialapplications.11,20,21 These sensors consist of a well-definedtip or wheel, or they utilize existent components, e.g., thefilling wire or the pressure wheel,22 for seam-tracking �Fig.
FIG. 1. Seam-tracking principle based on a scanning triangulation sensor.
FIG. 2. Seam-tracking principle using a line section sensor.
FIG. 3. Seam-tracking principle using gray-scale image processing.
2 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
4�. Pressure or elongation sensors at the mounting indicatedeviations of the joint and give feedback to the controller�active seam tracking�. Rare applications use the mechanicalguidance force of the joint edge onto a tracking wheel di-rectly to follow the joint.20 In this case, no controller is used�passive seam-tracking�.
Advantages of mechanical sensors are as follows:
• robust cheap setup;• small integrated sensor; existing components may be alien-
ated;• good dirt resistance, no optical parts.
Mechanical sensors are only usable for contoured jointslike fillet welds or Y-shaped butt welds; I-shaped butt weldsare not detectable. Furthermore, the joint has to be fairlylinear.17 A disadvantage is possible tip abrasion, which leadsto measurement errors. The tip also can damage the work-piece surface and abets collision risks.
B. Sensor arrangement
Two possible arrangements are used for seam tracking.In a less common setup, the sensor is installed fixed to the
robot hand. The welding head with the TCP �tool centerpoint; the point of incident of the welding laser� sits on atracking axis, which allows adjusting the TCP relative to therobot hand �Fig. 5�. The tracking axis is usually a preciseand fast linear axis, but can also be a rotary axis.23
In a more common setup the sensor is fixed to thewelding head �Figs. 6 and 7�. The advantage of this setup isthat the sensor moves with the TCP. Since the TCP matchesthe current joint position, the sensor is continuouslyreadjusted and therefore covers a greater deviation betweenjoint and robot trajectory.24 The tracking axis may be
FIG. 4. Mechanical seam-tracking principle using a tip, the filler wire, or atracking wheel.
FIG. 5. Seam-tracking setup with sensor fixed to the robot hand �open loopconfiguration�.
r robot
r TCP
sensor
TCP
trackingaxis
robothand
p corr
weldinghead
robot arm
joint
direction of travel
s
r jointx
y
FIG. 6. Seam-tracking setup with sensor fixed to the TCP �closed loopconfiguration�.
FIG. 7. Principle of seam-tracking systems.
3J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
abandoned and replaced by direct robot trajectoryadjustment; this corresponds to the latter setup type. Toreduce the sensor forerun and to reduce size, the sensor canbe integrated into the welding head observing the workpiece
coaxial to the laser beam.23,25
From the control point of view, the first setup type is anopen loop control, while the latter setup is closed loop. Dueto its obvious advantage of a larger acceptance width; thelatter principle is used more often; however, open loopcontrol is the more robust principle and provides easiersetup.
C. Control principle
Due to its more common use, the control principledescribed in the following refers to the closed loop controlconfiguration. The difference to the open loop configuration�Fig. 5� is that the joint position measured by the sensor sy isrelative to the actual TCP position r�TCP
s� = r�joint − r�TCP, �1�
whereas in open loop configuration it is relative to the robothand position r�robot;
s� = r�joint − r�robot. �2�
Since the TCP position, in comparison to the robot position,is controlled by the sensor measurand, this configuration isclosed loop.
Figure 8 illustrates the vector system used in thefollowing. The coordinate system r is fixed to the workpiece;the coordinate system p is fixed to the robot hand. paxis,y
determines the actual tracking axis position. The sensorcoordinate system s has its origin in the TCP; sx specifies thesensor forerun, sy the measured joint position relative to theTCP.
As a matter of principle, the sensor measures the jointposition not in the TCP position, but with a forerun oftypically 40–200 mm. Therefore, a time delay has to beconsidered between the measuring of a joint trajectorydeviation and its correction,
rtarget,y�t� = �rjoint,y�rjoint,x=rTCP,x�t� = sy�t − T� + rTCP,y�t − T� .
�3�
The delay depends on the forerun of the sensor sx and theaverage feed rate used to bridge the forerun distance. In thecase of a constant feed rate rrobot,x, the delay equals
T = sx/�rrobot,x�rrobot,x=const. �4�
The current TCP position in the global coordinate system rdepends on the robot hand and tracking axis position,
rTCP,y�t� = rrobot,y�t� + paxis,y�t� . �5�
The nominal position of the tracking axis pcorr,y equals thedifference of the target position rtarget,y and the robotposition rrobot,y. With formulas �3� and �5� it can becalculated to
pcorr,y�t� = rtarget,y�t� − rrobot,y�t�
= sy�t − T� + rrobot,y�t − T� + paxis,y�t − T�
− rrobot,y�t� . �6�
Usually, the lateral movement of the robot hand rrobot,y is notmeasurable and assumed to be zero within the accuracyrequirements. Without lateral robot movement thenominal axis position can be calculated by the sensormeasurand, the axis position and the time delay,
pcorr,y�t� = sy�t − T� + �paxis,y�t − T��rrobot,y=const. �7�
Figure 9 shows the simplified action diagram of the controlprinciple according to formula �7�.
This control principle requires that the current trackingaxis position paxis,y is measurable. If this is not possible, thenominal tracking axis position pcorr,y may be usedalternatively; however, the contouring error inducesadditional deviations. They can be reduced by using adynamic model of the tracking axis.26 This substitution canbe used in all control principles described in the following.
Simpler seam-tracking systems ignore the forerun anduse a conventional PID control principle to control thecorrection axis,27 which leads to principal positioning errors.Pritschow et al.28 quantify these errors in relation to TCPvelocity, sensor forerun, control parameters, and the jointtrajectory.
A seam-tracking concept from Thyssen Krupp AG25 alsoignores the forerun but claims that the residual error is notrelevant for the welding application due to a small sensordistance to the TCP of about 2 mm. This is true if the actualdelay of the controller matches the necessary delayaccording to formula �4�.
III. ERROR SOURCES IN SEAM-TRACKINGAPPLICATIONS
The seam-tracking control principle shown in the previ-ous chapter implicates the assumption of a constant linearrobot movement, requiring
• a constant feed rate;• no lateral movement of the robot hand;
FIG. 8. Vector definition, closed loop setup.
4 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255256
257
258
259
260
261
• no rotation of the welding head or robot hand; and• no movement or distortion of the workpiece.
In numerous real applications, these requirements are notgiven, thus leading to positioning errors, which will be ana-lyzed in the following.
A. Error type 1: Feed rate variation
As shown in formula �4�, the time delay T between themeasuring of a joint deviation and its correction depends onthe sensor forerun and the feed rate of the robot system. Thesensor forerun is given by the mechanical setup, but the feedrate eventually differs, provoked by
• 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 leads—in thecase of a nonlinear joint trajectory—to a positioning mis-match �Fig. 10�. Its length in feed direction can be deter-mined by
�feedrate,x = rtarget,x − rTCP,x = sx�1 −rrobot,x,real
rrobot,x,exp� , �8�
where rrobot,x,exp is the expected and rrobot,x,real is the realfeed rate. The absolute mismatch depends on the actualjoint contour.
B. Error type 2: Lateral robot movement
As shown in Fig. 9, the lateral movement of the robotrrobot,y is assumed to be zero in the control algorithm �notconsidering error correction techniques described later inthis article�.
However, in real applications lateral displacement doesoccur, for example caused by
• acceleration limitations of articulated arm robots;• deliberate lateral movement or rotation in 2D applications;• vibration; and• thermal distortion.
A hereby caused deviation is measured by the sensor but not
FIG. 9. Simplified seam-tracking control principle for closed loop configuration, not considering rotation and lateral robot movement.
FIG. 10. Positioning error caused by varied feed rate.
5J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
corrected instantly: due to the sensor forerun sx a time delayis considered, although instant correction would be required,cf. formula �6�.
The result is a temporary position mismatch, comparablewith a control delay. The length of the positioning mismatchequals the sensor forerun �Fig. 11�; its value equals the lat-eral movement of the robot,
�displacement,y = rtarget,y − rTCP,y = rrobot,y�t − T� − rrobot,y�t� .
�9�
C. Error type 3: Rotation of the welding head
Seam-tracking systems are extremely sensitive torotation of the welding head relative to the feed direction.The rotation provokes a lateral deviation of the TCPtrajectory and at the same time a shortening of the sensorforerun related to the feed direction,
r�TCP�t� = r�robot�t� + R · p�axis�t� , �10�
s��t� = R−1 · ��r�joint�rjoint,x=�r�TCP�t� + R · s��t��x− r�TCP,y�t�� , �11�
with
R = �cos � − sin �
sin � cos �� . �12�
Both modifications are not measured by the sensor andcause a constant position mismatch of the TCP �Fig. 12�.
FIG. 12. Deviation of the TCP position �rotation caused by welding headrotation �.
FIG. 13. Positioning error caused by rotation of the welding head.
FIG. 11. Positioning error caused by lateral displacement of the robotsystem.
6 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
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 feeddirection can be determined by
�rotation,x = rtarget,x − rTCP,x = sx�cos��� − 1� . �14�
D. Error type 4: Workpiece movement
A movement of the workpiece within the process timespan is usually caused by thermal distortion of theworkpiece or the fixture through local or global heat input.The movement can be parallel or lateral to the robotmovement or can induce rotation relative to the weldinghead, thus causing the same results as error types 1–3. Indifference to path deviations induced by the robot, aworkpiece movement cannot be corrected by conventionalerror correction techniques, which read the robot positioningdata and have no information about the workpiece position.
IV. ERROR CORRECTION TECHNIQUES
All errors described in the previous section are origi-nated in the sensor forerun with respect to the TCP. There-fore, a way to reduce sensor-originated deviation in all error
types is to abridge the sensor offset sx. A significant shorten-ing 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 fillerwire instead of an optical sensor for seam tracking. The fillerwire receives a force if abutted on the joint edge, which isgauged to control the correction axis. Since the tip of thefiller wire is coincident with the laser TCP, this principlepossesses zero sensor forerun, avoiding all previous de-scribed seam-tracking errors.
If a sensor forerun cannot be avoided, production ca-pable seam-tracking sensors use different correction tech-
TABLE I. Error correction techniques.
A: Dynamictime delay
B: Calibration onreference workpiece
C: Two-sensor positionmeasurements
D: Self-guidedprocessing
RequirementsFeed ratemeasurable
� �
High robot repeataccuracy
�
High robot timeaccuracy
�
Robot-sensorcalibration/sync
� � ���
Reference workpiecewith “ideal” contour
�
Corrected error1: Variation of feedrate
� �
2: Lateral robotmovement
� � �
3: Rotation ofwelding head
�
4: Workpiecemovement
�
trackingaxis
trackingsensor
robot
++ -
+
jointtrajectory 0
0
scal,y
yrobotr ,
( )xTCPyTCP rr ,,
( )xxTCPyjo srr +,int,
( )xxTCPy srs +,
Calibration vector(stored for work run)
yaxisp ,
(no correction)
(ideal contour)
FIG. 14. Storing of the calibration vector Scal�t� using a reference workpiecewith ideal joint trajectory �rjoint,y =0�.
7J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
niques to enable seam tracking for real world applications.Table I lists the most common correction techniques, theirrestrictions, and the error types they compensate.
Usually, method A �dynamic time delay� is combinedwith method B �calibration on reference workpiece� ormethod C �double joint position measuring�. If single errortypes can be precluded by mechanical assumptions or setuparrangements, individual error correction techniques may beobsolete.
Method D �self-guided processing� is capable to correctall four error types without needing a high robot accuracy,calibration or robot-sensor synchronization.
A. Dynamic time delay
If the actual feed rate is measurable—e.g., provided bythe robot control—it can be considered in the controlalgorithm by implementing a variable delay T� to correcterror type 1. The variable time delay T� has to meet therestriction
�t�=t−T�
t
rrobot,x�t�� = sx. �15�
B. Calibration on reference workpiece
A commonly used way to minimize error type 2 is togather a calibration vector Scal�t� along the welding path.24
Usually robots have a poor absolute accuracy but acomparatively good repeat accuracy. In a dry run, the sensormeasures the deviation to a reference workpiece, whichfeatures a joint deviation to the target path less or equal theneeded accuracy. If the sensor measures a deviation, thisdeviation is originated in a lateral movement of the robot�Fig. 14�.
In a work run the calibration value is time-equidistantsubtracted from the measurand to correct the robot pathdeviation �Fig. 15�,
rcorr,y�t� = sy�t − T� + rcorr,y�t − T� − scal,y�t� . �16�
The time-dependent sensor calibration eliminates errorscaused by lateral robot movement if an exact timing betweenthe robot movement and the sensor calibration and a goodrobot repeat accuracy can be ensured. However, the costs toproduce a reference workpiece and the effort of calibrationhave to be considered.
FIG. 15. Correction of error type 2 using a calibration vector Scal�t�.
FIG. 16. Correction of error type 2 using a second sensor s2 with a different forerun distance �s2,x�s1,x�.
8 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
C. Two-sensor position measurement
For correction of error type 2 without calibration on-linemeasuring of the lateral robot movement is required.Falldorf30 measures the lateral movement by calculating thedifference of the joint-to-TCP deviation at two unequalforerun distances s1x and s2,x �s2,x�s1,x� at the same time.The difference of the measurands at equal TCP positionsrTCP,x is originated in a lateral TCP movement rTCP,y,
rTCP,y�t� = rTCP,y�t − Ts� + s2,y�t − Ts� − s1,y�t� , �17�
with
Ts = �s2,x − s1,x�/rrobot,x. �18�
Knowing both the TCP and the joint trajectory, the lateralTCP position rTCP,y can be set onto the correspondinglateral joint position �Fig. 16�. This correction techniquerequires a known or constant feed rate and no weldinghead rotation.
D. Self-guided processing
The four error types described in Sec. IV can be allattributed to a lack of information. The seam-tracking sensoronly measures the joint position on the workpiece relative tothe current TCP position; however, the actual position of theTCP relative to the workpiece is not available. To calculatethe TCP trajectory with this incomplete information,irregularities besides deviation of the joint, particularlydeviation of the robot trajectory, welding head orientationand displacement of the workpiece, have to be assumed tobe not existent. The correction of the sensor offset has to becarried out time-based instead of position based; cf. formula�6�.
The lack of information is compensated with methods Band C by measuring an error value either by means of aseparate calibration run or a second joint positionmeasurement. Both methods can increase tracking accuracy;however, additional assumptions have to be assured �cf.Table I�.
For self-guided processing, the idea is to actuallymeasure the displacement between the TCP and theworkpiece. This additional information enables to calculateboth the TCP trajectory and the joint trajectory in acoordinate system stationary to the workpiece,
r�TCP�t� = r�const + �0
t
r�TCP�t� · dt , �19�
r�joint�t� = r�TCP�t� + s��t� . �20�
Knowing both the TCP and the joint trajectory, the lateralTCP position rTCP,y can be set onto the correspondinglateral joint position,
r�target�t� = �r�joint�i��rjoint,x�i�=rTCP,x�t�. �21�
The correction vector equals the vector difference,
p�corr�t� = r�target�t� + p�axis�t� − r�TCP�t� . �22�
With this concept, seam tracking is performed two-dimensional position based instead of time based. All fourerror types described in Sec. IV are inexistent.
The concept can be enhanced by a feedback of the targetposition to the robot control �dotted line in Fig. 17�. Withthis feedback, seam tracking of unknown joint trajectories isfeasible. In this case, the high accuracy positioning isperformed by the �two dimensional� tracking axis, while therough contour is followed by the robot.
V. REALIZATION OF A SELF-GUIDED PROCESSINGHEAD
A. Displacement sensor
A sensor principle to measure the relative displacementto a workpiece surface is described in Ref. 34. It utilizes therough surface finish of the workpiece: the surface structureis unique and stationary on the workpiece. A relativedisplacement between the sensor and the workpiecetherefore can be measured by finding the maximum crosscorrelation of areas in consecutive observed images �Fig.18�, a principle also known as optical flow.32 The relativevelocity corresponds to the displacement normalized by theframe 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 forerunand secure installation. The workpiece surface is illuminatedcoaxially to the camera with a low power illumination laser�Fig. 19�.
The camera observes an area of approximately6�6 mm2. Using narrow band filters and a small and highintense illumination spot, the thermal radiation of thewelding process and back reflection of the laser beam can besuppressed completely �Fig. 20�.
trackingaxis
trackingsensor
robot
++
+
++
jointtrajectory
displacementsensor
-+
( ) ( )( ) ( )trirjointtarget
xTCPxjointirtr
,, ==rr
robotrr
TCPrr
axisprjointrr
jointsr
TCPr&r′
TCPr ′r
jointr ′r
targetr ′r
corrpr
FIG. 17. Correction of all four error types using an additional sensor to trackthe relative displacement to the workpiece �self-guided processing�.
9J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
B. Joint position measurement
For joint measurement, the same images are utilized asfor the displacement measurement. The technique of jointrecognition by gray-scale image shape analysis is wellknown.18 If necessary, light section measurement could beintegrated using an additional illumination. However, thegray-scale image processing developed for this applicationturned out to be stable at different materials as stainlesssteel, mild steel, copper, and aluminium alloys with punchedas well as laser cut edges.
C. Results
The sensor has been adapted to a 6 kW CO2 scannersystem �scan field 40�40 mm2� with a forerun to the TCPof 200 mm �noncoaxial setup�. The motion system is anarticulated robot with comparatively low path accuracy.Tests where performed on mild steel with a sinus-formedbutt joint �Fig. 21�.
The system is able to follow the joint with randomlyvarying feed rates independent of the welding headorientation and the robot movement. Due to performancelimitations of the control system, the maximum feed rate ofthe test setup is limited to 5 m /min.
D. Further development
The self-guided processing optic is currently enhancedto real two-dimensional seam-tracking by integrating thesensor as well a 2D positioning system �scanner ormotorized adjustable mirror� into the optical path of the laserbeam �coaxial setup, Fig. 22�.
This setup allows to follow a two-dimensional jointwithout rotating the welding head. The self-guided opticonly needs to be moved within the scanner field upon thejoint and the laser beam “finds its way.” Even hand guidedor vehicle hooked up laser welding can be realized forbutt-welding and other high accuracy applications.
Another field of current research is the analysis ofadditional information of the coaxial camera sensor forfurther process information �melt pool geometry, splatter,and seam geometry�, which enables seam tracking andprocess monitoring with one single sensor system. Thecoaxial illumination of the process zone allows robustfeature detection.33,34
VI. CONCLUSION
Seam-tracking sensors are useful tools for laser welding,if the workpiece or fixture accuracy does not meet the needed
FIG. 18. Measurement of relative displacement between consecutive images by finding the maximum cross correlation of areas in consecutive observedimages.
10 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
restrictions. However, due to the necessary sensor forerun,conventional control principles based on PID control havelimited accuracy.
These tracking deviations can be eliminated using anadvanced control principle, which takes the forerun into ac-count. Nowadays, this is state of the art in most commer-cially available seam-tracking systems. Yet, additional devia-tions occur through deviation in the feed rate, a lateralmovement of the robot hand, rotation of the processing head,or movement or distortion of the workpiece. Different errorcorrection techniques are used to reduce these deviations;however, they all limit flexibility, are time consuming, andneed advanced interfacing with the robot system, thus in-
creasing setup and running costs. Also, standard articulatedrobots cannot maintain the required path accuracy and re-peatability restrictions, therefore more expensive and lessflexible gantry systems, tricept robots, or optimized articu-lated robots have to be used.35,36
To eliminate the identified error sources, a seam-trackingprinciple and device has been developed, which computes anerror-free correction vector independent of the robot andworkpiece movement. It consists of a multifunctional sensor,which additionally to the seam position measures the dis-placement between the processing head and the workpiece.This information is used by an advanced control principle todetermine the trajectory of the joint as well as the TCP in aglobal coordinate system. Therefore, seam tracking can beperformed two-dimensional position-based instead of one-dimensional time based.
FIG. 19. Optical setup of the combined displacement and seam-tracking sensor �left: stand-alone sensor and right: coaxial setup�.
FIG. 20. Observed workpiece surface and welding process �coaxial setup�. FIG. 21. Self-guided seam tracking of a sine-wave formed joint.
11J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
A concept is introduced, which combines a coaxial mea-surement concept with a laser positioning system. A process-ing head using this concept is capable to follow joints ortraces on the workpiece self-guided, which means indepen-dent of the actual movement of the processing head andwithout any prior calibration or information of the trajectory.The absolute position accuracy is independent from the ac-curacy of the motion system, which enables applicationswith low accuracy robots or even hand moved systems.
Nomenclature
paxis,x � distance of the TCP to the robot hand centerpoint in feed direction
paxis,y � lateral position of the TCP relative to therobot hand center point �actual tracking axisposition�
pcorr,y � joint position lateral to TCP relative to therobot hand center point �desired trackingaxis position�
r�TCP � tool center point in global coordinate systemr�joint � joint position lateral to sensor position in
global coordinate systemr�robot � robot hand center point in global coordinate
systemrrobot,x � feed rate �of the robot hand�rrobot,y � lateral movement of the robot handr�target � joint position lateral to TCP in global coor-
dinate systemSx � 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 robothand
T � time delay to compensate sensor forerun
�feedrate,x � deviation between desired and actual TCPposition caused by feed rate variation in feeddirection
�displacement,y � lateral deviation between desired and actualTCP position caused by lateral movement ofthe robot hand
�rotation,x � deviation between desired and actual TCPposition caused by rotation of the weldinghead in feed direction
�rotation,y � lateral deviation between desired and actualTCP position caused by rotation of the weld-ing head
1M. Schultz, “Fertigungsqualität beim 3D-Laserstrahlschweißen vonBlechformbauteilen,” Dissertation, University Erlangen-Nuernberg, Ger-many, 1997, p. 58.
2F. Lange and G. Hirzinger, “Adaptive minimization of the maximal pathdeviations of industrial robots,” Proceedings of the European ControlConference, Karlsruhe, VDI/VDE GMA, 1999.
3F. Garnich, “IWB-Forschungsberichte,” Laserbearbeitung mit Robotern�Springer, Berlin, 1992�, p. 50.
4P. Hoffmann, “Verfahrensfolge Laserstrahlschneiden und-schweißen,”Fertigungstechnik Erlangen �C. Hanser, Muenchen, 1992�.
5M. Loo, Y. Hamidieh, and V. Milenkovic, “Generic path control for robotapplications,” Robots 14 Conference Proceedings �Society of Manufac-turing Engineers, Detroit, 1990�, 1–16.
6G. Pritschow, M. Bauder, and A. Horn, “Erhöhung der Bahngenauigkeitvon Industrierobotern,” Robotersysteme 8, 162–170 �1992�.
7A. Albu-Schäffer, “Regelung von Robotern mit elastischen Gelenken amBeispiel der DLR-Leichtbauarme,” 2002.
8A. Reek, “Strategien zur Fokuspositionierung beim Laserstrahlsch-weißen,” Ph.D. thesis, H. Utz Verlag, München, 2000.
9L. Weiss, A. Sanderson, and C. Neuman, “Dynamic sensor-based controlof robots with visual feedback,” IEEE J. Rob. Autom. 3, 404–417 �1987�.
10G. Starke, “Nahtführungssensor zur adaptiven Steuerung vonHandhabungseinrichtungen zum Lichtbogenschweißen,” Dissertation,University RWTH Aachen, 1983.
11K. Fuchs, Flexible, sensorgesteuerte Roboterschweißsysteme, Aachen,Fotodruck Mainz, 1987.
12H. Bögel, “Laser führt Schweißroboter mit Pilotsensor,” Laser-Praxis 1,67–68 �1990�.
13W. Trunzer, H. Lindl, and H. Schwarz, “Einsatz eines voll 3D-fähigenschnellen Sensorsystems zur Nahtverfolgung beim Laserstrahlsch-weißen,” Laser in der Technik, edited by W. Waidelich �Springer, Berlin,1993�, pp. 405–410.
14Falldorf Sensor GmbH, Falldorf Sensor, 2007 �http://www.falldorf-sensor.de�.
15ServoRobot Inc., Seam finding, Seam Tracking ROBO-TRAC, 2007�http://www.servorobot.com/english/Manufacturing_solutions/robo-trac.htm�.
16Precitec KG, Laser Welding-Monitoring-Optical Seam Tracking UsingLight Stripe Sensor, 2007 �http://www.precitec.de/precitec/enxml/products/lpf.html�.
17Meta-Scout GmbH, SCOUT Nahtverfolgung, 2007 �http://www.scout-sensor.com/�.
18A. Horn, “Optische Sensorik zur Bahnführung von Industrierobotern mithohen Bahngeschwindigkeiten,” ISW Forschung und Praxis �Springer,Berlin, 1994�, p. 103.
19Plasmo Industrietechnik GmbH, Plasmo Industrietechnik GmbH, 2007�http://www.plasmo.eu�.
20I. Haschke, “Verfahren und Vorrichtung zum Fügen von Werkstückteilenmittels eines Energiestrahls, insbesondere Laserstrahls,” Patent No.DE10006852 �August 26, 2004�.
21T. Jäckel, F. Hanschmann, and S. Schiwy, “Process and device for joiningat least two workpieces with laser beam and filter material, the latter usedas contact element for guiding along the joining line,” European PatentNo. 06022685 �June 27, 2007�.
22HIGHYAG Lasertechnologie GmbH, 2007 �www.highyag.de�.23Permanova Lasersystem AB, 2007 �www.permanova.se�.24J.-P. Boillot, ’Robot feature tracking devices and methods,” U.S. Patent
No. 6,430,472 �December 20, 1999�.
FIG. 22. Coaxial setup of camera sensor, illumination, and laser positioningsystem.
12 J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
543
544
545
546
547
548
549
550
551
552
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
579
580
581
582
583
584
585
586
587
588
589
590
591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648
25Thyssenkrupp Steel AG, F. Behr, E. Blumensaat, C. Dornscheidt, M.Koch, J. Plha, S. Wischmann, L. Ott, andA. Schäfer, Patent ApplicationNo. WO/2007/088122 �August 9, 2007�.
26R. Kram, U. Kreienkamp, and R. Friedrich, “Robotersteuerungsfunk-tionen und Schnittstellen zur Ankopplung vorlaufender Seneoren,” Intel-ligente Sensorsysteme der Fertigungstechnik, edited by J. Rogos�Springer, Berlin, 1989�, pp. 117–125.
27G. Gruhler, “Sensorgeführte Programmierung bahngesteuerter Industri-eroboter,” SW Forschung und Praxis �Springer, Berlin, 1987�, p. 67.
28G. Pritschow and A. Horn, “Dynamik derzeitiger Sensorregelkreise fürIndustrieroboter,” Robotersysteme 7, 178–184 �1991�.
29J. Müller-Borhanian, “Integration optischer Messmethoden zur Proz-esskontrolle beim Laserstrahlschweissen �INESS�,” Abschlussbericht zumVerbundprojekt �Forschungs- und Tagungsberichte, München, 2005�.
30H. Falldorf, “Nahtverfolgung für das Laserstrahlschweissen unter beson-derer Berücksichtigung eines vom Schweissautomaten unabhängigen Sys-tems,” Ph.D. thesis, Fortschritt-Berichte VDI. Reihe 8, Mess-,Steuerungs-und Regelungstechnik, 1995.
31B. Regaard, S. Kaierle, W. Schulz, and A. Moalem, “Advantages of ex-ternal illumination for monitoring and control of laser materials process-
ing,” Proceedings of ICALEO 2005, Miami. FL, October 31–November3, 2005, pp. 915–919.
32S. S. Beauchemin and J. L. Barron, “The computation of optical flow,”ACM Comput. Surv. 27, 433–466 �1995�.
33W. F. Clocksin, J. S. E. Bromley, P. G. Davey, C. G. Morgan, and A. R.Vidler, “An Implementation of Model-Based Visual Feedback for RobotArc Welding of Thin Sheet Steel,” Int. J. Robot. Res. 4, 13–26 �1985�.
34J. Gedicke, B. Regaard, A. Gillner, and S. Kaierle, “Kontrolle beim Mik-roschweißen: Automatisierte Prozessüberwachung durch koaxiale Proz-esskontrolle mit Fremdbeleuchtung,” Laser-Technik-Journal 3, 33–37�2006�.
35B. Regaard, S. Kaierle, and R. Poprawe, “Error detection in lap weldingapplications using on-line melt pool contour analysis by coaxial processmonitoring with external illumination,” Lasers in Manufacturing 2007:Proceedings of the Fourth International WLT-Conference Lasers in Manu-facturing �2007�, pp. 471–476.
36A. Huwer, “Sensorsystem zur Erfassung variabler Fügespaltweiten beimLaserstrahlschweißen im Stumpfstoß,” Aachener Berichte Fügetechnik�Shaker, Aachen, 1993�.
13J. Laser Appl., Vol. 21, No. 4, November 2009 Regaard, Kaierle, and Poprawe
649
650651652653654655656657658659660661662663664665666667668
669
670671672673674675676677678679680681682683684685686687