Seam Tracking for High Precision Laser Welding

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.


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.


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.


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�.


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�.


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�.


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.


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.


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.


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�.


649

650651652653654655656657658659660661662663664665666667668

669

670671672673674675676677678679680681682683684685686687

Documents

Seam Tracking for High Precision Laser Welding