JOURNAL OF LASER APPLICATIONS
VOLUME 21, NUMBER 4
NOVEMBER 2009
Seam-tracking for high precision laser welding applications—Methods, 2 restrictions and enhanced concepts 1
Boris Regaard,1,a兲 Stefan Kaierle,2,b兲 and Reinhart Poprawe2
3 4 5
1
Fraunhofer Center for Laser Technology, Plymouth, Michigan 48170 Fraunhofer Institute for Laser Technology, D-52074 Aachen, Germany
2
6
共Received 17 October 2008; accepted for publication 20 August 2009兲
7
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 共typically 50 m兲 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 is—dependent on 40 the controller algorithm—within 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
a兲
Electronic mail:
[email protected] Electronic mail:
[email protected]
b兲
1042-346X/2009/21共4兲/1/0/$25.00
1
algorithms, and mechanical improvements;2,5–7 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 and—after calibration—repetitious 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 1980’s.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 共sensor兲, 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
2
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 共linear or rotary兲, 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 共Fig. 1兲. 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 共line scan camera, 89 one-dimensional search兲; 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 共point94 shaped laser beam兲. 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 共Fig. 2兲. The detector 102 has two-dimensional resolution 共CCD or CMOS camera兲. 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 共dependent on the cam- 105 era framerate and image processing algorithm兲; 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 共camera兲 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 共or brightness兲 within the camera image 共Fig. 3兲. Advantages of this concept are as follows:
108
• small sensor design 共no triangulation angle needed兲; • simultaneous measurements in different distances possible; higher measurement reliability; • detection of thin butt joints 共technical zero gap兲 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 3°–7°, dependent on the surface finish and material. Seam-tracking sensors using this principle are available 共e.g. Plasmo Position controller19兲. 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 共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
3
direction of travel joint sensor
s
TCP welding head
p
corr
tracking axis
FIG. 4. Mechanical seam-tracking principle using a tip, the filler wire, or a tracking wheel.
r
4兲. Pressure or elongation sensors at the mounting indicate 147 deviations of the joint and give feedback to the controller 148 共active seam tracking兲. 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 共passive seam-tracking兲. 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
r
joint
r
robot
robot arm
FIG. 6. Seam-tracking setup with sensor fixed to the TCP 共closed loop configuration兲.
robot hand. The welding head with the TCP 共tool center point; the point of incident of the welding laser兲 sits on a tracking axis, which allows adjusting the TCP relative to the robot hand 共Fig. 5兲. 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 共Figs. 6 and 7兲. 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 共open loop configuration兲.
FIG. 7. Principle of seam-tracking systems.
166 167 168 169 170 171 172 173 174 175 176
4
J. Laser Appl., Vol. 21, No. 4, November 2009
Regaard, Kaierle, and Poprawe
rtarget,y共t兲 = 兩rjoint,y兩rjoint,x=rTCP,x共t兲 = sy共t − T兲 + rTCP,y共t − T兲. 共3兲 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 r˙robot,x, the delay equals T = sx/兩r˙robot,x兩r˙robot,x=const .
共4兲 217
The current TCP position in the global coordinate system r 218 219 depends on the robot hand and tracking axis position, rTCP,y共t兲 = rrobot,y共t兲 + paxis,y共t兲.
共5兲 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 共3兲 and 共5兲 it can be 223 224 calculated to
FIG. 8. Vector definition, closed loop setup.
pcorr,y共t兲 = rtarget,y共t兲 − rrobot,y共t兲 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 共Fig. 5兲 is that the joint position measured by the sensor s y is 193 relative to the actual TCP position rជTCP 189
194 195 196 197
sជ = rជjoint − rជTCP ,
共1兲
whereas in open loop configuration it is relative to the robot hand position rជrobot; sជ = rជjoint − rជrobot .
共2兲
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
= sy共t − T兲 + rrobot,y共t − T兲 + paxis,y共t − T兲 − rrobot,y共t兲.
226
共6兲 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,y共t兲 = sy共t − T兲 + 兩paxis,y共t − T兲兩rrobot,y=const .
228 229 230 231 232
共7兲 233
Figure 9 shows the simplified action diagram of the control principle according to formula 共7兲. 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 共4兲.
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
5
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 共4兲, 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 leads—in the case of a nonlinear joint trajectory—to a positioning mismatch 共Fig. 10兲. Its length in feed direction can be determined by
冉
feedrate,x = rtarget,x − rTCP,x = sx 1 −
冊
r˙robot,x,real , r˙robot,x,exp
共8兲
where r˙robot,x,exp is the expected and r˙robot,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 r˙robot,y is assumed to be zero in the control algorithm 共not 288 considering error correction techniques described later in 289 this article兲.
FIG. 10. Positioning error caused by varied feed rate.
6
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 共6兲. 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 共Fig. 11兲; its value equals the lat303 eral movement of the robot, 298
displacement,y = rtarget,y − rTCP,y = rrobot,y共t − T兲 − rrobot,y共t兲. 共9兲 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
rជTCP共t兲 = rជrobot共t兲 + R · pជ axis共t兲,
共10兲 311
sជ共t兲 = R−1 · 共兩rជjoint兩rjoint,x=共rជTCP共t兲 + R · sជ共t兲兲x − rជTCP,y共t兲兲,
共11兲 312
with
R=
FIG. 12. Deviation of the TCP position rotation caused by welding head rotation ␣.
313
冉
cos ␣ − sin ␣ sin ␣
cos ␣
冊
.
共12兲
314
Both modifications are not measured by the sensor and 315 cause a constant position mismatch of the TCP 共Fig. 12兲. 316
J. Laser Appl., Vol. 21, No. 4, November 2009
Regaard, Kaierle, and Poprawe
7
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 共Fig. 13兲 corresponds to rotation,y = rtarget,y − rTCP,y = sx sin共␣兲,
共13兲
where ␣ is the welding head rotation. The shortening in feed direction can be determined by rotation,x = rtarget,x − rTCP,x = sx共cos共␣兲 − 1兲.
共14兲
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 1–3. 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
0
+
(no correction)
+
rTCP , y (rTCP , x )
s y (rTCP , x + s x )
0
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
共⫻兲
⫻
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 Scal共t兲 using a reference workpiece with ideal joint trajectory 共rjoint,y = 0兲.
338 339 340 341 342 343 344 345 346 347 348 349
8
J. Laser Appl., Vol. 21, No. 4, November 2009
Regaard, Kaierle, and Poprawe
A. Dynamic time delay
362
If the actual feed rate is measurable—e.g., provided by the robot control—it 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
冕
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 共dynamic time delay兲 is combined 354 with method B 共calibration on reference workpiece兲 or 355 method C 共double joint position measuring兲. 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 共self-guided processing兲 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
t⬘=t−T⬘
FIG. 15. Correction of error type 2 using a calibration vector Scal共t兲.
364
r˙robot,x共t⬘兲 = sx .
共15兲 368
B. Calibration on reference workpiece
369
A commonly used way to minimize error type 2 is to gather a calibration vector Scal共t兲 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 共Fig. 14兲. In a work run the calibration value is time-equidistant subtracted from the measurand to correct the robot path deviation 共Fig. 15兲,
370 371
rcorr,y共t兲 = sy共t − T兲 + rcorr,y共t − T兲 − scal,y共t兲.
373 374 375 376 377 378 379 380 381
共16兲 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 共s2,x ⬎ s1,x兲.
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
pជ corr共t兲 = rជtarget共t兲 + pជ axis共t兲 − rជTCP共t兲.
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 共s2,x ⬎ s1,x兲 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,y共t兲 = rTCP,y共t − Ts兲 + s2,y共t − Ts兲 − s1,y共t兲,
共17兲
with Ts = 共s2,x − s1,x兲/r˙robot,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 共Fig. 16兲. 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 共6兲. 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 共cf. 423 Table I兲. 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
rជTCP共t兲 = rជconst +
冕
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 共dotted line in Fig. 17兲. With this feedback, seam tracking of unknown joint trajectories is feasible. In this case, the high accuracy positioning is performed by the 共two dimensional兲 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 共Fig. 18兲, 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 共CMOS camera兲, 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 共Fig. 19兲. 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 共Fig. 20兲.
449 450
joint trajectory
robot r rrobot
tracking axis
r paxis
+ +
共19兲
0
r r joint
r rTCP
+
displacement sensor
r pcorr
t
rជ˙ TCP共t兲 · dt,
431 432 433 434 435
rជjoint共t兲 = rជTCP共t兲 + sជ共t兲.
共20兲
Knowing both the TCP and the joint trajectory, the lateral TCP position rTCP,y can be set onto the corresponding lateral joint position, rជtarget共t兲 = 兩rជjoint共i兲兩rjoint,x共i兲=rTCP,x共t兲 . The correction vector equals the vector difference,
共21兲
r s joint
r ′ rtarget r ′ r joint
r r rtarget (t ) = r joint (i )
tracking sensor
r ′ r&TCP
+
430
共22兲 436 438 439 440 441 442 443 444 445
共18兲
400
429
9
r ′ 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 共self-guided processing兲.
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 共scan field 40⫻ 40 mm2兲 with a forerun to the TCP of 200 mm 共noncoaxial setup兲. The motion system is an articulated robot with comparatively low path accuracy. Tests where performed on mild steel with a sinus-formed butt joint 共Fig. 21兲. 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 共scanner or motorized adjustable mirror兲 into the optical path of the laser beam 共coaxial setup, Fig. 22兲. 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 共melt pool geometry, splatter, and seam geometry兲, 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 共left: stand-alone sensor and right: coaxial setup兲.
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 共coaxial setup兲.
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 共actual tracking axis position兲 pcorr,y ⫽ joint position lateral to TCP relative to the robot hand center point 共desired tracking axis position兲 rជTCP ⫽ tool center point in global coordinate system rជjoint ⫽ joint position lateral to sensor position in global coordinate system rជrobot ⫽ robot hand center point in global coordinate system r˙robot,x ⫽ feed rate 共of the robot hand兲 r˙robot,y ⫽ lateral movement of the robot hand rជtarget ⫽ joint position lateral to TCP in global coordinate system Sx ⫽ forerun of the sensor to TCP 共closed loop兲 or robot hand 共open loop兲 Sy ⫽ lateral position of the joint relative to the sensor 共at sensor position兲 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, “Fertigungsqualität beim 3D-Laserstrahlschweißen 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 共Springer, Berlin, 1992兲, p. 50. 4 P. Hoffmann, “Verfahrensfolge Laserstrahlschneiden und-schweißen,” Fertigungstechnik Erlangen 共C. Hanser, Muenchen, 1992兲. 5 M. Loo, Y. Hamidieh, and V. Milenkovic, “Generic path control for robot applications,” Robots 14 Conference Proceedings 共Society of Manufacturing Engineers, Detroit, 1990兲, 1–16. 6 G. Pritschow, M. Bauder, and A. Horn, “Erhöhung der Bahngenauigkeit von Industrierobotern,” Robotersysteme 8, 162–170 共1992兲. 7 A. Albu-Schäffer, “Regelung von Robotern mit elastischen Gelenken am Beispiel der DLR-Leichtbauarme,” 2002. 8 A. Reek, “Strategien zur Fokuspositionierung beim Laserstrahlschweißen,” Ph.D. thesis, H. Utz Verlag, München, 2000. 9 L. Weiss, A. Sanderson, and C. Neuman, “Dynamic sensor-based control of robots with visual feedback,” IEEE J. Rob. Autom. 3, 404–417 共1987兲. 10 G. Starke, “Nahtführungssensor zur adaptiven Steuerung von Handhabungseinrichtungen zum Lichtbogenschweißen,” Dissertation, University RWTH Aachen, 1983. 11 K. Fuchs, Flexible, sensorgesteuerte Roboterschweißsysteme, Aachen, Fotodruck Mainz, 1987. 12 H. Bögel, “Laser führt Schweißroboter mit Pilotsensor,” Laser-Praxis 1, 67–68 共1990兲. 13 W. Trunzer, H. Lindl, and H. Schwarz, “Einsatz eines voll 3D-fähigen schnellen Sensorsystems zur Nahtverfolgung beim Laserstrahlschweißen,” Laser in der Technik, edited by W. Waidelich 共Springer, Berlin, 1993兲, pp. 405–410. 14 Falldorf Sensor GmbH, Falldorf Sensor, 2007 共http://www.falldorfsensor.de兲. 15 ServoRobot Inc., Seam finding, Seam Tracking ROBO-TRAC, 2007 共http://www.servorobot.com/english/Manufacturing_solutions/robotrac.htm兲. 16 Precitec KG, Laser Welding-Monitoring-Optical Seam Tracking Using Light Stripe Sensor, 2007 共http://www.precitec.de/precitec/enxml/ products/lpf.html兲. 17 Meta-Scout GmbH, SCOUT Nahtverfolgung, 2007 共http://www.scoutsensor.com/兲. 18 A. Horn, “Optische Sensorik zur Bahnführung von Industrierobotern mit hohen Bahngeschwindigkeiten,” ISW Forschung und Praxis 共Springer, Berlin, 1994兲, p. 103. 19 Plasmo Industrietechnik GmbH, Plasmo Industrietechnik GmbH, 2007 共http://www.plasmo.eu兲. 20 I. Haschke, “Verfahren und Vorrichtung zum Fügen von Werkstückteilen mittels eines Energiestrahls, insbesondere Laserstrahls,” Patent No. DE10006852 共August 26, 2004兲. 21 T. Jäckel, 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 共June 27, 2007兲. 22 HIGHYAG Lasertechnologie GmbH, 2007 共www.highyag.de兲. 23 Permanova Lasersystem AB, 2007 共www.permanova.se兲. 24 J.-P. Boillot, ’Robot feature tracking devices and methods,” U.S. Patent No. 6,430,472 共December 20, 1999兲.
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. Schäfer, Patent Application No. WO/2007/088122 共August 9, 2007兲. 26 R. Kram, U. Kreienkamp, and R. Friedrich, “Robotersteuerungsfunktionen und Schnittstellen zur Ankopplung vorlaufender Seneoren,” Intelligente Sensorsysteme der Fertigungstechnik, edited by J. Rogos 共Springer, Berlin, 1989兲, pp. 117–125. 27 G. Gruhler, “Sensorgeführte Programmierung bahngesteuerter Industrieroboter,” SW Forschung und Praxis 共Springer, Berlin, 1987兲, p. 67. 28 G. Pritschow and A. Horn, “Dynamik derzeitiger Sensorregelkreise für Industrieroboter,” Robotersysteme 7, 178–184 共1991兲. 29 J. Müller-Borhanian, “Integration optischer Messmethoden zur Prozesskontrolle beim Laserstrahlschweissen 共INESS兲,” Abschlussbericht zum Verbundprojekt 共Forschungs- und Tagungsberichte, München, 2005兲. 30 H. Falldorf, “Nahtverfolgung für das Laserstrahlschweissen unter besonderer Berücksichtigung eines vom Schweissautomaten unabhängigen 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 31–November 3, 2005, pp. 915–919. 32 S. S. Beauchemin and J. L. Barron, “The computation of optical flow,” ACM Comput. Surv. 27, 433–466 共1995兲. 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, 13–26 共1985兲. 34 J. Gedicke, B. Regaard, A. Gillner, and S. Kaierle, “Kontrolle beim Mikroschweißen: Automatisierte Prozessüberwachung durch koaxiale Prozesskontrolle mit Fremdbeleuchtung,” Laser-Technik-Journal 3, 33–37 共2006兲. 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 共2007兲, pp. 471–476. 36 A. Huwer, “Sensorsystem zur Erfassung variabler Fügespaltweiten beim Laserstrahlschweißen im Stumpfstoß,” Aachener Berichte Fügetechnik 共Shaker, Aachen, 1993兲.
669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
