Robotic seam tracking

The principles of contact and non-contact sensors for robotic seam tracking are reviewed. Non-contacting sensors are based on either an electromagnetic or an acoustic principle. Electromagnetic sensors in the lovi frequency regime induce eddy currents which can be used for the detection of surface discontinuities. In th( high frequency regime (visual spectrum), laser light stripe systems in connection with vision cameras can b( successfully used for seam location. Other optical systems and infrared thermography are briefly described Alternatively, acoustic systems using time of flight measurements can be applied to the characterization o: seams. Finally, through the arc systems and their limitations for seam tracking are outlined. The need fo] off-line programming techniques for pathplanning with seam tracking devices is emphasized. based scam tracking and part location systems has become caicial ever since industrial robots have been usec in processes like arc welding, glueing or sealing. Around the world, researchers are competing to be the firs to produce a general purpose sensor which can be used for adaptive control in robotic seam tracking. Thus far, several contacting and non-contacting seam tracking devices have been developed, and a few ar< commercially available. However, none of these seam tracking systems can be shown to meet al requirements of the arc welding users. Numerous lists of the requirements which a seam tracking device mus fulfill were established in the past The authors consider the following points as important: • General purpose: The same system should be usable for different weld geometries. • Ruggedness: The system must be functional in an industrial environment • The system should be usable in connection with different arc welding techniques. • The system should operate in real time: Position and weld parameters can be adapted to the welding situation. • The system should provide three dimensional information on the seam and fit up. • Sensing devices should be relatively small in size so as not to limit the motion of the arc welding torch. • The sensory system should be inexpensive in comparison with the total cost of the robotic arc welding system. 2. Seam Tracking Devices Seam tracking systems presently being investigated may be divided into two major categories: tactile am non-tactile. Tactile probes are attractive because of their simplicity and reliability [1]. The touch sensor usually consist! of one or more LVDTs. Sets of strain gauges mounted on a probe which can elastically deflect are als< common. However, in using tactile probes it is difficult if not impossible, to provide information on the join fit up. Furthermore, certain weld geometries, e.g., welding into corners, exclude the use of contacting sensors, In addition to seam tracking, tactile sensors may provide very useful information on overall part location Let us assume that the parts to be welded together have some simple, well defined geometrical features suet as locally plane regions or circular cross section, etc. In these cases, the use of tactile probes can reduce the complexity of real time sensors. The welding of pipe joints is one example where simple touch, e.g., electrica contact between the welding wire and the workpiece surface, allows one to determine the geometry of the pipe intersections and automatically generate a robot welding program. However, if poor fit up or randon deviations of the joint intersection prohibit this simple approach, a combination of a part location sensor, e.g. simple touch, with a real time seam tracking system may provide the highest degree of flexibility. The toucl: sensor detects systematic deviations from the pre-programmed position. The seam tracking system correct! random deviations along the seam trajectory. there are only two ways in which information can be transmitted in a non-contact fashion: by acoustic wavei or electromagnetic waves. There is a wide spectrum of frequencies available in which non-contact sensor may operate. A sensor emitting electromagnetic waves at a frequency between 10 to 100 khz would typically be referred to as a magnetic sensor. Frequencies of approximately 100 Thz have been used in laser light strips systems, weld pool vision systems, laser scanners and systems based on arrays of photodiodes. Attempts hav< also been reported on the use of infrared (10 Thz) cameras looking directly at the weld pool. There is a basic difference in how geometric information is obtained through acoustic and electromagnetic waves. This is du< to the large difference between the speed of sound and speed of light. Acoustic sensors can easily be used ir connection with time of flight measurements, i.e., measuring the time difference between emitting anc receiving an acoustic pulse. On the other hand, optical systems operating at a distance of a few centimeter! must use triangulation schemes to compute the distance between the sensor and the surface of the object. Another possible non-contacting method for seam tracking is measuring changes in the welding current oi arc voltage. Obviously, this seam tracking technique can only be used with arc welding, and as later discussec in this paper, only if certain geometric conditions are met In the following sections, we discuss currently used seam tracking systems and the principles on which thej are based. 3. Vision and Structured Light A common way of characterizing joint location and fit up in arc welding applications is depicted in Fig. 1 A so-called structured light source, a pattern of lines or a grid, is projected at a certain angle onto the surface of the object. The angle between the optical axis of the camera and the light source is constant. The camera's two dimensional image of the projected light stripe principally allows a three dimensional reconstruction oi the joint geometry. Significant progress was made, over the last few years, in the area of image acquisition and analysis. SRI [2] has developed a procedure for constructing weld joint models based on vision data. The true joint location i< determined by comparing a model with the original joint prototype specified by the user. This system has been successfully used to control the motion of a robot in welding low carbon, hot rolled steel plates. Although the light stripe approach seems very attractive, it has several drawbacks. The physical dimension of the camera and the light source normally moving ahead of the welding torch may prevent the torch from accessing walls and corners. Also, the distance between the light stripe projected on the object and the welding torch limits the radius of trajectories which can be negotiated with this system. Furthermore, the continuous exposure of both camera and light source to the welding environment, dust and spatter, results in reduced light intensity and decreased resolution. Interference of the bright arc with the illumination system represents a serious shielding or noise filtering problem. Both image filtering and three dimensional characterization by triangulation require significant computational support. The latter point is not so much a technical constraint, but an economic one. The noise due to the arc can be overcome by using a so-called (wo pass system. First, the camera performs a search cycle locating and characterizing the seam. The information obtained is then used to adjust the lucuuwu jji/iiiia aiv pivpvuy avjju^Lvu. i nv avuiui wjfv.iv vein uv v-amvu uui ui u. uiuwii H1511V1 jpvvu uiuu ui\ welding cycle. In many manufacturing applications, the search cycle does not represent a serious through-pu degradation. The disadvantage of this approach is primarily that local distortions during welding cannot b< taken into account. These distortions are caused by thermal expansions of the material. It should be noted that as an alternative to projecting a stationary light pattern onto the surface of the joints the surface can be periodically scanned with a focused laser beam. Conceptually, the three dimensiona surface pattern will be reconstructed in a way similar to that done with the light stripe. 4. Weldpooi Vision All previously described systems can only be applied to torch guidance. None can be applied to arc procesi control. Based on a concept originally proposed by Richardson [3], GE has built a vision system in whicl changes in the weld puddle and the joint are obtained as viewed coaxially with the torch electrode. Puddl< geometry is thus monitored, and parameters like puddle size can be used to control welding parameters Furthermore, a joint tracking system based on structured light is incorporated in the GE weld vision system The system, as demonstrated during the last AWS show, Philadelphia 1983, could be applied to TIG weldinj with a speed of 6 inches per minute. Extensions of that system to MIG, which require much higher weldinj speeds, were reported under development 5. Arrays of Photodiodes A robust seam tracking system, as shown in Fig. 2, was developed by P. Drews [4] at The University o: Aachen, W. Germany. In this tracking system, a halogen light source is transmitted via fiber optics to th< sensor system. (In the halogen light frequency range, the light intensity of the welding arc was shown to be minimal.) An elliptical light spot is projected onto the joint, and a linear array of 256 photodiodes measure! the reflected light intensity. The recorded intensity pattern is then related to the actual geometry of the welc joint. The seam geometry is found by determining the relative intensity normal to the seam trajectory. Lowej density indicates a longer distance between the sensor and the surface of the workpiece. 6. Infrared The information content of infrared light emitted from the weld pool has been investigated [5]. The result* showed that arc misalignment, groove geometry faults, variations in penetration and impurities could, ir principle, be detected by infrared thermography. While such systems are likely to have long term impact or adaptive welding control, significant amount of research still needs to be done to fully expedite their potentia for industrial applications. 7. Magnetic Fields The alteration of magnetic fields by induced eddy currents in electrically contacting media is a simpk mechanism by which surface discontinuities, i.e., edges or gaps, can be detected. Fig. 3 shows the geometry o\ a sensor which can be used for seam tracking applications. Without the presence of the workpiece, twe symmetric alternating magnetic fields would be induced. Naturally, the same potential is induced in the sens* coils underneath the inducer coils. The presence of a conducting medium (as indicated in Fig. 3) will cause the induction of eddy currents which in turn tend to modify the originally symmetric flux field. According tc asymmetries in the workpiece, different potentials are induced in the two sense coils. This difference provide a simple means for locating surface discontinuities. A prototype of the described system was developed a Carnegie-Mellon University [6]. Extensions of that system into arrays of inducers and sensors are currentl; under development. This system provides and excellent tool for providing measurements of the relativ< position of gaps or edges in the surface to the sensor. Absolute distance information is possible only if th< magnitude of the induced signals is calibrated with respect to a specific material. Another drawback is th< need to keep the sensor relatively close to the workpiece (less than 1/2). 8. Acoustic Waves Acoustic waves can be used to perform a three dimensional characterization of workpiece surfaces. T< extract range information, acoustic systems typically use a pulse technique originally developed by Pellam anc Gait [7], The sensing system utilizes a piezoelectric transducer to convert electrical energy into ultrasoni< energy, and vice versa. The same sensor, therefore, acts as emitter and receiver. The attenuation of th< ultrasonic wave by the medium of propagation constrains the choice of the transducer's resonant frequency Ultlrasonic waves propagated through air are attenuated in proportion to the square of the frequency Consequently, the resonant frequency should be chosen as low as possible. To overcome some of th< attenuating losses, the transducer's primary transmission surface is spherically concave, resulting in a focusec ultrasonic wave. Other considerations suggest choosing a high operating frequency. High frequencies produce narrowei radiation patterns and lead to better resolution when sampling the distance from the transducer to a point or the workpiece surface. By performing a periodic two dimensional sampling of the distance to the workpiece, moving the sensoi back and forth as indicated in Fig. 4, a complete surface characterization can be performed, including sean position. A major problem encountered in the surface characterization process is caused by the limitec aperture of the acoustic sensor. In a prototype system developed at Carnegie-Mellon University, the sensor*! line of sight must not deviate from the surface normal at the sampling point by more than approximately i degrees (sensor frequency 1 Mhz). Otherwise, the amplitude of the received echo is insufficient to provid< reliable range information. However, knowing approximately the local surface orientation of the workpiece which is a reasonable assumption in a manufacturing environment, the sensor can be maneuvered into ifc angular region of proper operation. Another inherent problem with acoustic waves is the temperatun dependence of the speed of sound. In single pass welding, this problem is largely overcome by shielding th( sensor from the arc and real time speed of sound calibration by using a reference transducer. More serious problems are expected in multipass welding applications. Alternatively to time of flight measurements, it is possible to monitor the interference pattern between the emitted and the reflected waves. The interference pattern can be sampled by arrays of microphones ai specific locations around the object [8]. Changes in the object's shape will result in changes in the interference pattern. While this system was originally designed for inspection of parts, its principle appears applicable foi robotic seam tracking. Extensive measurements in the past have shown that the average welding current or average arc voltage, is Droportional to the distance between the electrode and the work piece [9]. Hence, executing a weaving notion with the welding torch normal to the seam trajectory reveals the surface profile of the joint. Weaving 3f the torch may be accomplished by simple mechanical oscillation or by having the plasma oscillate in an alternating magnetic field. Higher oscillation frequencies, and consequently a more precise seam :haracterization, can be accomplished with the alternating magnetic field. However, this approach maj involve space intrusion problems since the magnetic field generator must be mounted relatively close to the torch tip. Two control algorithms for through the arc sensing have been investigated: template matching and the :onceptually more simple differential control. Template matching assumes the availability of a template signal. This signal is expressed as a function of the displacement with respect to the center of the arc weave pattern. The torch must be guided so that the integrated difference between the template signal and the actual signal is minimized. The difference control strategy simply tries to minimize the difference between the signals obtained at bott extremes of the oscillating motion. Similar to the horizontal control strategy, the vertical distance between the torch and the workpiece can b< controlled by comparing the signal as measured at the center of the oscillation to a predetermined vertica current reference. Through the arc sensing has been successfully tried with* a number of different arc welding processe: including TIG, MIG, fluxed-core and Submerged arc welding. Major drawbacks of through the arc sensing result from the fact that the dimensions of the joint mus exceed some critical dimension, e.g., through the arc sensing today is not applicable to sheet metal welding Another principal problem is that a signal can be obtained only after the arc has been established. This mean that this method cannot be used for finding the starting point of the weld. Most manufacturers of arc welding robots offer the integration of through the arc systems into their robo controller. 10. Adaptive Control Strategies for Robotic Seam Tracking Most experiments to date involving real time seam tracking allow at most three degrees of freedom o adaptive change. These degrees of freedom typically refer to the workpiece frame rather than the joint frani of the robot. Ideally, one likes to adjust both position and orientation, i.e., six degrees of freedom, of the toe mounted at the end effector of the robot. More research needs to be directed towards the development o general purpose algorithms which can be executed on today's microprocessors. A recent analysis [10] using an inverse Jacobian approach showed that a full six dimensional adaptiv ilie Jacobian is the differential of the manipulator transform. A complication in using preview sensors stems from the need to know exactly the transform relating th< tool (torch) and the sensor. Since the chances of misalignment in a manufacturing environment are great automatic calibration procedures will become mandatory. Finally, the availability of seam tracking sensors will reduce the robot programming effort since fewe: points along the seam trajectory will need to be pre-programmed. On the other hand, the coordinated motibi of both sensor and the welding torch, while avoiding interference with obstacles like fixtures, is likely tc increase the overall complexity of robot programming. Further complications are due to programming o recovery strategies in the event the sensor fails to obtain a healthy reading. The recovery strategies will diffe: for different applications and might even require user input. In view of these complications, it is assumed tha techniques such as off-line path planning and off-line programming will become vital steps in the successfii implementation of robots with seam tracking systems. 11. Summary The principles of several seam tracking devices have been discussed. The majority of these devices are stil in the trial stage, and very little industrial experience with regard to these seam tracking sensors is presentlj available. Therefore, it is not possible to say what system will ultimately succeed; most likely, a combinatior of a through the arc system with an appropriate preview sensor will prove the most satisfactory. 12. Acknowledgements This work was supported by the Welding Consortium, The Robotics Institute, Carnegie-Mellon University Pittsburgh, PA. All the aforementioned sensors, except through the arc, are called preview sensors since they locate the seam ahead of the torch.